Search

There are examples of how unconscious bias can influence actions of people. In the judiciary, however, despite some examples there is no general theory on whether different demographic attributes such as gender, seniority or ethnicity affect case sentencing. We aim to gain insight into this issue by analyzing over 100k decisions of three different areas of law with the goal of understanding whether judge identity or judge attributes such as gender and seniority can be inferred from decision documents. We find that stylistic features of decisions are predictive of judge identities, their gender and their seniority, a finding that is aligned with results from analysis of written texts outside the judiciary. Surprisingly, we find that features based on legislation cited are also predictive of judge identities and attributes. While own content reuse by judges can explain our ability to predict judge identities, no specific reduced set of features can explain the differences we find in the legislation cited of decisions when we group judges by gender or seniority. Our findings open the door for further research on how these differences translate into how judges apply the law and, ultimately, to promote a more transparent and fair judiciary system. [ABSTRACT FROM AUTHOR]

This paper identifies trends in the application of big data in the transport sector and categorizes research work across scientific subfields. The systematic analysis considered literature published between 2012 and 2022. A total of 2671 studies were evaluated from a dataset of 3532 collected papers, and bibliometric techniques were applied to capture the evolution of research interest over the years and identify the most influential studies. The proposed unsupervised classification model defined categories and classified the relevant articles based on their particular scientific interest using representative keywords from the title, abstract, and keywords (referred to as top words). The model's performance was verified with an accuracy of 91% using Naïve Bayesian and Convolutional Neural Networks approach. The analysis identified eight research topics, with urban transport planning and smart city applications being the dominant categories. This paper contributes to the literature by proposing a methodology for literature analysis, identifying emerging scientific areas, and highlighting potential directions for future research. [ABSTRACT FROM AUTHOR]

The Internet of Things (IoT) devices spawn growing diverse social platforms and online data at the network edge, propelling the development of cross-platform applications. To integrate cross-platform data, user identity linkage is envisioned as a promising technique by detecting whether different accounts from multiple social networks belong to the same identity. The profile and social relationship information of IoT users may be inconsistent, which deteriorates the reliability of the effectiveness of identity linkage. To this end, we propose a topic and knowledge-enhanced model for edge-enabled IoT user identity linkage across social networks, named TKM, which conducts feature representation of user generated contents from both post-level and account-level for identity linkage. Specifically, a topic-enhanced method is designed to extract features at the post-level. Meanwhile, we develop an external knowledge-based Siamese neural network for user-generated content alignment at the account-level. Finally, we show the superiority of TKM over existing methods on two real-world datasets. The results demonstrate the improvement in prediction and retrieval performance achieved by utilizing both post-level and account-level representation for identity linkage across social networks. [ABSTRACT FROM AUTHOR]

The scholarly literature at the intersection of sales and ethics is vast and, therefore, difficult to summarize. To explore the state of the sales–ethics landscape, the authors apply probabilistic topic modeling to a dataset composed of 293 journal articles published from 1980 to 2022. The critical examination of the results leads to a framework that identifies 32 topics and groups these topics into five high-level topic areas. Building on these topics and topic areas, the authors explore where future research on sales and ethics should focus, using in-depth interviews with 30 scholars and 15 practitioners. The results of these interviews reveal important implications for sales and ethics and overarching research questions regarding (1) understanding existing realities, (2) understanding new realities, and (3) advancing research practices. In doing so, this study provides a platform for much-needed research and scholarly discourse on sales and ethics. [ABSTRACT FROM AUTHOR]

Unlike text analysis for which topic models are historically developed, traffic video analysis is dealing with much simpler topics, made of restricted motion patterns. In this paper, we propose a dual-layer direction-oriented framework for more efficient traffic motion patterns description based on topic models through considering the simplicity of traffic topics. The aforesaid framework compels the involved topic models to learn the foreknown visually meaningful motion patterns that exist in traffic scenes, as developed theoretically in this paper. Experimental results produced by common datasets show that the proposed method provides more intuitive topics for traffic flow description. Based on experimental results, our framework outperforms other topic-model based methods by 4% to more than 11% in detecting abnormal events, in terms of the area under the Receiver Operating Characteristic curve. In addition to that, in a scene analysis evaluation at intersections equipped with traffic signals, our method reaches 4% higher traffic phase detection accuracy, compared to conventional topic models. [ABSTRACT FROM AUTHOR]

As an extension to the standard symmetric latent Dirichlet allocation topic model, we implement asymmetric Beta-Liouville as a conjugate prior to the multinomial and therefore propose the maximum a posteriori for latent Beta-Liouville allocation as an alternative to maximum likelihood estimator for models such as probabilistic latent semantic indexing, unigrams, and mixture of unigrams. Since most Bayesian posteriors, for complex models, are intractable in general, we propose a point estimate (the mode) that offers a much tractable solution. The maximum a posteriori hypotheses using point estimates are much easier than full Bayesian analysis that integrates over the entire parameter space. We show that the proposed maximum a posteriori reduces the three-level hierarchical latent Beta-Liouville allocation to two-level topic mixture as we marginalize out the latent variables. In each document, the maximum a posteriori provides a soft assignment and constructs dense expectation–maximization probabilities over each word (responsibilities) for accurate estimates. For simplicity, we present a stochastic at word-level online expectation–maximization algorithm as an optimization method for maximum a posteriori latent Beta-Liouville allocation estimation whose unnormalized reparameterization is equivalent to a stochastic collapsed variational Bayes. This implicit connection between the collapsed space and expectation–maximization-based maximum a posteriori latent Beta-Liouville allocation shows its flexibility and helps in providing alternative to model selection. We characterize efficiency in the proposed approach for its ability to simultaneously stream both large-scale data and parameters seamlessly. The performance of the model using predictive perplexities as evaluation method shows the robustness of the proposed technique with text document datasets. [ABSTRACT FROM AUTHOR]

Urbanization is crucially important for people to improve the quality of life. Thus, it is of importance to study the evolution of hot topics to explore the functions of cities for meeting the increasing demands of people. In this paper, we explored the semantic analysis of hot topics and trends in urban studies from the literature, which provides a research direction for future studies. Based on articles collected from the Science Citation Index Expanded and Conference Proceedings Citation Index-Science databases from 2000 to 2016, we found that the number of urban studies increased in stability during that time. Followed by England and China, USA was the largest contributor for studies in this field. Based on the keywords and abstracts of these articles, we extracted the topics of the study using a clustering method and topic model, and calculated the hot values of the topics. Finally, we obtained 15 hot topics in the field of urban studies, among which "city", "school", "regional economic", and "estate" were the hottest topics that indicated the focus of the research study. An anomaly detection method was used to analyze the change trend of topics' hot values, and we found that the hot value of these topics overall were on the rise, especially "urban education" and "urban planning" increased significantly, which indicated that they attracted an increasing amount of scholars' attention, but the hot value of "health" and "Gis" decreased significantly recently, which suggested that research interest in these two topics is decreasing. [ABSTRACT FROM AUTHOR]

Recently, research on short text topic models has addressed the challenges of social media datasets. These models are typically evaluated using automated measures. However, recent work suggests that these evaluation measures do not inform whether the topics produced can yield meaningful insights for those examining social media data. Efforts to address this issue, including gauging the alignment between automated and human evaluation tasks, are hampered by a lack of knowledge about how researchers use topic models. Further problems could arise if researchers do not construct topic models optimally or use them in a way that exceeds the models' limitations. These scenarios threaten the validity of topic model development and the insights produced by researchers employing topic modelling as a methodology. However, there is currently a lack of information about how and why topic models are used in applied research. As such, we performed a systematic literature review of 189 articles where topic modelling was used for social media analysis to understand how and why topic models are used for social media analysis. Our results suggest that the development of topic models is not aligned with the needs of those who use them for social media analysis. We have found that researchers use topic models sub-optimally. There is a lack of methodological support for researchers to build and interpret topics. We offer a set of recommendations for topic model researchers to address these problems and bridge the gap between development and applied research on short text topic models. [ABSTRACT FROM AUTHOR]

Nowadays, online users write short messages to share their feelings on social networking sites, such as discussion forums, question answering websites, etc., making these sites very popular. The increase in these short-term messages causes a huge data sparsity, making emotion recognition a challenging task. Therefore, a word co-occurrence pattern called biterms is generated from a large-scale dataset to prevent severe data sparsity issues. The topic modeling algorithms and acceleration algorithms are implemented to extract more reliable topics from the group of terms. Based on biterm technique, in this paper, a new algorithm called "Affected biterm emotion topic" is proposed for emotion recognition from a short text. For the experimental purpose, two popular short text datasets, SemEval and International Survey on Emotion Antecedents and Reactions (ISEAR), are used to investigate the performance of the proposed algorithm with the benchmark methods light latent dirichlet allocation (LLDA), biterm topic model (BTM), emotion-topic model (ETM), contextual sentiment topic model (CSTM), Sentiment latent topic model (SLTM) and siasme network based supervised topic model (SNSTM). The proposed algorithm is evaluated using the benchmark methods for mean, variance, and accuracy. The experimental result shows that the proposed algorithm is effective in analyzing emotions. [ABSTRACT FROM AUTHOR]

In this paper, the SASC (Sentiment Analysis based on Sentiment Clustering) method is proposed to solve the problems of low accuracy and poor stability in the review sentiment clustering methods. Through two-stage sentiment clustering, the hidden sentiment information among the review texts is obtained to improve the accuracy and stability of the results. Specifically, in the first stage, the review representation vector construction method is put forward through the topic model LDA. Then the second stage uses K-means algorithm to achieve further optimization of the sentiment clustering results. In the experiment part, the evaluation methods of sentiment clustering are firstly introduced, and then a series of experiments are carried out on two widely used datasets Large Movie Review Dataset v1.0 and Multi-Domain Sentiment Dataset. Experiment results indicate that compared with other methods, the SASC method proposed in this paper has better clustering accuracy and stability. [ABSTRACT FROM AUTHOR]

The rapid transition of livestock husbandry in the 20th century involved a broad adoption of slurry-based livestock housing systems that resulted in farm economic benefits, but also in societal debate related to the environment and animal welfare. In this article, we apply the method of topic modeling to four major German newspapers to identify thematic emphases and changes in coverage around "slurry". We considered more than 2300 articles published between 1971 and 2020. Our results show that reporting encompasses economic, environmental, and social topics in which slurry is represented mostly critically ("poisonous substance"), occasionally neutrally ("scent of countryside"), or rarely positively ("input for the bioeconomy"). Three meta-themes overarch the majority of issues and reflect public discourse on agriculture: (i) the dichotomy of agricultural industrialization and family farming; (ii) contrasting actualities of factory farming and animal welfare; and (iii) the responsibility of policy for the emergence, existence and solution of livestock and slurry-related problems. A more balanced recognition of mutual values and constraints by the media could contribute to a discursive reconciliation of public and private interests. [ABSTRACT FROM AUTHOR]

Neural networks, primarily recurrent and convolutional Neural networks, have been proven successful in text classification. However, convolutional models could be limited when classification tasks are determined by long-range semantic dependency. While the recurrent ones can capture long-range dependency, the sequential architecture of which could constrain the training speed. Meanwhile, traditional networks encode the entire document in a single pass, which omits the hierarchical structure of the document. To address the above issues, this study presents T-HMAN, a Topic-aware Hierarchical Multiple Attention Network for text classification. A multi-head self-attention coupled with convolutional filters is developed to capture long-range dependency via integrating the convolution features from each attention head. Meanwhile, T-HMAN combines topic distributions generated by Latent Dirichlet Allocation (LDA) with sentence-level and document-level inputs respectively in a hierarchical architecture. The proposed model surpasses the accuracies of the current state-of-the-art hierarchical models on five publicly accessible datasets. The ablation study demonstrates that the involvement of multiple attention mechanisms brings significant improvement. The current topic distributions are fixed vectors generated by LDA, the topic distributions will be parameterized and updated simultaneously with the model weights in future work. [ABSTRACT FROM AUTHOR]

Laws and legal decision-making regulate how societies function. Therefore, they evolve and adapt to new social paradigms and reflect changes in culture and social norms, and are a good proxy for the evolution of socially sensitive issues. Here, we use an information-theoretic methodology to quantitatively track trends and shifts in the evolution of large corpora of judicial decisions, and thus to detect periods in which disruptive topics arise. When applied to a large database containing the full text of over 100,000 judicial decisions from Spanish courts, we are able to identify an abrupt change in housing-related decisions around 2016. Because our information-theoretic approach pinpoints the specific content that drives change, we are also able to interpret the results in terms of the role played by legislative changes, landmark decisions, and the influence of social movements. [ABSTRACT FROM AUTHOR]

Every epidemic affects the real lives of many people around the world and leads to terrible consequences. Recently, many tweets about the COVID-19 pandemic have been shared publicly on social media platforms. The analysis of these tweets is helpful for emergency response organizations to prioritize their tasks and make better decisions. However, most of these tweets are non-informative, which is a challenge for establishing an automated system to detect useful information in social media. Furthermore, existing methods ignore unlabeled data and topic background knowledge, which can provide additional semantic information. In this paper, we propose a novel Topic-Aware BERT (TABERT) model to solve the above challenges. TABERT first leverages a topic model to extract the latent topics of tweets. Secondly, a flexible framework is used to combine topic information with the output of BERT. Finally, we adopt adversarial training to achieve semi-supervised learning, and a large amount of unlabeled data can be used to improve inner representations of the model. Experimental results on the dataset of COVID-19 English tweets show that our model outperforms classic and state-of-the-art baselines. [ABSTRACT FROM AUTHOR]

The problem of constructing a correct topic model is relevant for the automatic processing of large collections of text messages tasks. This paper considers an approach to the topic categorization assessing for a collection of short text messages (labeled up by experts) based on the clustering of terms that make up the message text. The results of a computer experiment on clustering a set of terms are presented and discussed. As part of the experiment, a graph was constructed that reflects the relations and some numerical characteristics of the terms that form the topic model for a collection of messages. Analysis of the structure of the graph allows one to generate some practical recommendations for reorganizing the topic model presented in the expert labeled text collection. [ABSTRACT FROM AUTHOR]

Modern software development involves multiple developers working remotely in a distributed manner around the world. Software bugs are continuously generated for multiple reasons across various modules. It is possible that one software bug can affect multiple modules, and there can be multiple developers associated with it. Furthermore, many software bug reports are unlabeled, vague, and noisy. The triager faces significant challenges in identifying multiple causes of software bugs and finding expert developers for bug fixing. In this paper, the fuzzy set is extended to Intuitionistic Fuzzy Sets (IFS), and a novel bug triaging approach based on Intuitionistic Fuzzy Similarity (IFSim) measures is presented to overcome the aforementioned problems. The topic model is used to discover multiple relationships between developers and software bugs. IFS is used to separate developers based on their degree of membership and non-membership in a particular software category, with a degree of hesitation for some developers. For a new bug, 15 different IFSim measure techniques are investigated to compute the similarity with the existing software bugs. Finally, a fuzzy α -cut is applied to find expert developers to repair it. The best results are obtained by considering the number of topics of 15 and 12 taxonomic terms for each topic. Among all the IFSim measure techniques, the similarity techniques proposed by Ye outperform other techniques. Experiments are carried out on available benchmark data sets, and the results are compared to traditional machine learning algorithms and the fuzzy logic-based Bugzie model. [ABSTRACT FROM AUTHOR]

Global warming has caused the Arctic Ocean ice cover to shrink. This endangers the environment but has made traversing the Arctic channel possible. Therefore, the strategic position of the Arctic has been significantly improved. As a near-Arctic country, China has formulated relevant policies that will be directly impacted by changes in the international relations between the eight Arctic countries (regions). A comprehensive and real-time analysis of the various characteristics of the Arctic geographical relationship is required in China, which helps formulate political, economic, and diplomatic countermeasures. Massive global real-time open databases provide news data from major media in various countries. This makes it possible to monitor geographical relationships in real-time. This paper explores key elements of the social development of eight Arctic countries (regions) over 2013–2019 based on the GDELT database and the method of labeled latent Dirichlet allocation. This paper also constructs the national interaction network and identifies the evolution pattern for the relationships between Arctic countries (regions). The following conclusions are drawn. (1) Arctic news hotspot is now focusing on climate change/ice cap melting which is becoming the main driving factor for changes in geographical relationships in the Arctic. (2) There is a strong correlation between the number of news pieces about ice cap melting and the sea ice area. (3) With the melting of the ice caps, the social, economic, and military activities in the Arctic have been booming, and the competition for dominance is becoming increasingly fierce. In general, there is a pattern of domination by Russia and Canada. [ABSTRACT FROM AUTHOR]

The detection of emerging trends is of great interest to many stakeholders such as government and industry. Previous research focused on the machine learning, network analysis and time series analysis based on the bibliometrics data and made a promising progress. However, these approaches inevitably have time delay problems. For the reason that leader papers of "emerging topics" share the similar characters with the "cover papers", this study present a novel approach to translate the "emerging topics" detection to "cover paper" prediction. By using "AdaBoost model" and topic model, we construct a machine learning framework to imitate the top journal (chief) editor's judgement to select cover paper from material science. The results of our prediction were validated by consulting with field experts. This approach was also suitable for the Nature, Science, and Cell journals. [ABSTRACT FROM AUTHOR]

Individual passenger travel patterns have significant value in understanding passenger's behavior, such as learning the hidden clusters of locations, time, and passengers. The learned clusters further enable commercially beneficial actions such as customized services, promotions, data-driven urban-use planning, peak hour discovery, and so on. However, the individualized passenger modeling is very challenging for the following reasons: 1) The individual passenger travel data are multi-dimensional spatiotemporal big data, including at least the origin, destination, and time dimensions; 2) Moreover, individualized passenger travel patterns usually depend on the external environment, such as the distances and functions of locations, which are ignored in most current works. This work proposes a multi-clustering model to learn the latent clusters along the multiple dimensions of Origin, Destination, Time, and eventually, Passenger (ODT-P). We develop a graph-regularized tensor Latent Dirichlet Allocation (LDA) model by first extending the traditional LDA model into a tensor version and then applies to individual travel data. Then, the external information of stations is formulated as semantic graphs and incorporated as the Laplacian regularizations; Furthermore, to improve the model scalability when dealing with massive data, an online stochastic learning method based on tensorized variational Expectation-Maximization algorithm is developed. Finally, a case study based on passengers in the Hong Kong metro system is conducted and demonstrates that a better clustering performance is achieved compared to state-of-the-arts with the improvement in point-wise mutual information index and algorithm convergence speed by a factor of two. [ABSTRACT FROM AUTHOR]

Topic model is a well proven tool to investigate the semantic content of textual corpora. Yet corpora sometimes include texts in several languages, making it impossible to apply language-specific computational approaches over their entire content. This is the problem we encountered when setting to analyze a philosophy of science corpus spanning over eight decades and including original articles in Dutch, German and French, on top of a large majority of articles in English. To circumvent this multilingual problem, we use machine-translation tools to bulk translate non-English documents into English. Though largely imperfect, especially syntactically, these translations nevertheless provide correctly translated terms and preserve the semantic proximity of documents with respect to one another. To assess the quality of this translation step, we develop a “semantic topology preservation test” that relies on estimating the extent to which document-to-document distances have been preserved during translation. We then conduct an LDA topic-model analysis over the entire corpus of translated and English original texts, and compare it to a topic-model done over the English original texts only. We thereby identify the specific contribution of the translated texts. These studies reveal a more complete picture of main topics that can found in the philosophy of science literature, especially during the early days of the discipline when numerous articles were published in languages other than English. [ABSTRACT FROM AUTHOR]

Copyright of Frontiers of Information Technology & Electronic Engineering is the property of Springer Nature and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

At present, we live in prosperity contrary to the past times. As income increases, people enjoy wealth, but more people tend to pursue their own inner happiness: travel. People go to other places or visit foreign countries for business or journey. This study aims to identify the best tour route for foreign tourists in South Korea. Based on the review analysis results, this paper also aims to put forward techniques and methodologies required in practical affairs when developing a travel site or a travel application. On this note, it collected tourists' reviews at the Tripadvisor official website and conducted text mining technique as well as network analysis using R and Tagxedo, which is a big data analytic tool. The analysis results displayed that there were differences in travel preference, and especially, individual travelers had difficulty traveling by public transportation and selecting travel destinations. Therefore, customized travel routes were suggested for convenient use among travelers. [ABSTRACT FROM AUTHOR]

Recently, the proposed non-parametric Bayesian based techniques which aim to model short-length textual documents through the multinomial distribution on the bag-of-words (BOW), aka mixture model-based approach. Although existing model can effectively deal with the topic/concept drift and textual sparsity problems, they are unable to exploit the semantic sequential representation of text as well as the co-occurrence relationships between words. To meet these challenges, we propose a novel approach called as GOWSeqStream. Our proposed model is a joint integration of graph-of-words (GOW) and deep sequential encoding within the Dirichlet Process Mixture Model (DPMM) framework to improve the performance of text clustering task. Extensive experiments in benchmark real-world datasets demonstrate the effectiveness of our proposed GOWSeqStream model in comparing with recent state-of-the-art baselines. Experimental outputs in terms of NMI standard metric demonstrate the outperformances of proposed GOWSeqStream model over the recent well-known text stream clustering baselines, such as MStream, NPMM and OSDM. [ABSTRACT FROM AUTHOR]

The combination of the topic model and the semantic method can help to discover the semantic distributions of topics and the changing characteristics of the semantic distributions, further providing a new perspective for the research of topic evolution. This study proposes a solution for quantifying the semantic distributions and the changing characteristics based on words in topic evolution through the Dynamic topic model (DTM) and the word2vec model. A dataset in the field of Library and information science (LIS) is utilized in the empirical study, and the topic-semantic probability distribution is derived. The evolving dynamics of the topics are constructed. The characteristics of evolving dynamics are used to explain the semantic distributions of topics in topic evolution. Then, the regularities of evolving dynamics are summarized to explain the changing characteristics of semantic distributions in topic evolution. Results show that no topic is distributed in a single semantic concept, and most topics correspond to various semantic concepts in LIS. The three kinds of topics in LIS are the convergent, diffusive, and stable topics. The discovery of different modes of topic evolution can further prove the development of the field. In addition, findings indicate that the popularity of topics and the characteristics of evolving dynamics of topics are irrelevant. [ABSTRACT FROM AUTHOR]

Deciding when and which products to recommend to whom is always an essential issue for retailers. In this study, we propose a mixed framework with two components to capture customer buying behavior and its changes over time and visualize these results to better help retailers choose and target products strategically for marketing. In this framework, a topic model is first used to extract customer's purchase behavior instead of association rules or K-means as mainly used in market field. To automatically choose the optimal number of topics, we implement an approach proposed by Koltcov et al. on point-of-sale (POS) data in the supermarket. Meanwhile, to grasp the change of topics over time, we divided monthly POS data in half and applied the topic model with Renyi entropy separately. The results suggest that splitting data might be a better way to understand customer behavior. Second, we consider how to develop an effective way to visualize the results of the topic model, which is essential, because in a supermarket context, simply knowing which product categories are included under which topics is not enough to support how a supermarket promotes their products. To address this, we design a three-layer visualization approach to better interpret the topic model results and to help retailers design target promotion strategies. The design of visualization was overlooked by studies related to the use of topic models on supermarket data. Finally, to demonstrate the usefulness of our proposed framework, we conduct a simple scenario-based analysis between our framework and other models, such as Latent Dirichlet Allocation (LDA) and the Dynamic Topic Model (DTM). The results show that for most periods, our proposed framework outperforms LDA and DTM. [ABSTRACT FROM AUTHOR]

This study investigates how major media outlets, particularly CNN and Fox News, presented the "Unite the Right Rally" in Charlottesville Virginia. A computer-assisted content analysis was conducted on 1-week online news articles about the rally from the two networks. Evidence from Topic Modeling and Sentiment analysis suggests that CNN and Fox News share similarities in reporting the rally regardless of their target audiences. The findings demonstrate the efforts made by the media to uphold color-blindness ideology, underplay racial conflict, and subtly normalize white supremacy. The main contribution of this paper lies in (a) offering empirical evidence of media effect, (b) extending the examination of media effect from traditional forms to the online platform, (c) reexamining the political leaning the media posits, and (d) bridging media study and color-blind racism. [ABSTRACT FROM AUTHOR]

Deep learning has developed rapidly in recent years, attracting the attention of numerous researchers. Since a wide range of topics are covered in this field, we are wondering what topics researchers have concerned about. However, after investigation, we find that very few researchers have paid attention to this demand. In this paper, we conduct a large-scale study to analyze the questions faced by deep learning developers. We use Stack Overflow, one of the largest question&answer sites, as our data source, and extract 32969 posts about deep learning as our studied dataset. After filtering, augmenting and pre-processing the post datasets from Stack Overflow, we use the latent Dirichlet allocation (LDA) topic model to summarize 30 topics based on their text content. In addition, we measure the difficulty and popularity of each topic, compare the different issues faced by different deep learning frameworks, and analyze the development trend of each topic. Our main results are as follows: (1) developers ask a broad spectrum of questions about deep-learning, ranging from Data Shape to Object Detection; (2) Gradient Propagation is the most popular among all the topics and (3) Object Detection is the most difficult; (4) issues of Package Installation, Code Understanding and Method Introduction are common in the current different deep learning frameworks; (5) there are three trends in these topics, e.g., a significant rising trend is found in the number of discussion on Data Shape. Finally, based on our research findings, we make some targeted and valuable suggestions for developers, researchers, educators, and framework providers of deep learning. [ABSTRACT FROM AUTHOR]

Currently available Collaborative Filtering(CF) algorithms often utilize user behavior data to generate recommendations. The similarity calculation between users is mostly based on the scores, without considering the explicit attributes of the users with profiles, as these are difficult to generate, or their evolution of preferences over time. This paper proposes a collaborative filtering algorithm named T-LDA (Time-decay Dirichlet Allocation), which is based on the topic model. In this method, we generate a hybrid score for similarity calculation with topic model. However, most topic models ignore the attribute of time order. In order to further improve the prediction accuracy, a time-decay function is introduced in topic model. The experimental results show that this algorithm has better performance than currently available algorithms on the MovieLens dataset, Netflix dataset and la.fm dataset. [ABSTRACT FROM AUTHOR]

We exploit variations in access rules on the European Court of Justice to explore the effect of procedural inclusiveness on the agenda of international adjudicators. Using natural language processing methods, we analyze the entire universe of ECJ decisions up to 2015, mapping issue prevalence across time, procedure and litigant type. We find evidence that the more inclusive annulment and referral procedures are associated with greater issue heterogeneity whereas less inclusive infringement procedure displays greater issue cohesiveness as well as greater issue stability over time. [ABSTRACT FROM AUTHOR]

The hierarchical Dirichlet processes (HDP) topic model is a Bayesian nonparametric model that provides a flexible mixed-membership to documents through topic allocation to each word. In this paper, we consider dynamic HDP topic models, in which the generative model changes in time, and develop a novel algorithm to update the posterior distribution dynamically by combining the variational inference algorithm and the power prior approach. An important advantage of the proposed algorithm is that it updates the posterior distribution by reusing a given batch algorithm without specifying a complicated dynamic generative model. Thus the proposed algorithm is conceptually and computationally simpler. By analyzing real datasets, we show that the proposed algorithm is a useful alternative approach to dynamic HDP topic identification. [ABSTRACT FROM AUTHOR]

Inferring trip destination in smart card data with only tap-in control is an important application. Most existing methods estimate trip destinations based on the continuity of trip chains, while the destinations of isolated/unlinked trips cannot be properly handled. We address this problem with a probabilistic topic model. A three-dimensional latent dirichlet allocation model is developed to extract latent topics of departure time, origin, and destination among the population; each passenger's travel behavior is characterized by a latent topic distribution defined on a three-dimensional simplex. Given the origin station and departure time, the most likely destination can be obtained by statistical inference. Furthermore, we propose to represent stations by their rank of visiting frequency, which transforms divergent spatial patterns into similar behavioral regularities. The proposed destination estimation framework is tested on Guangzhou Metro smart card data, in which the ground-truth is available. Compared with benchmark models, the topic model not only shows increased accuracy but also captures essential latent patterns in passengers' travel behavior. The proposed topic model can be used to infer the destination of unlinked trips, analyze travel patterns, and passenger clustering. [ABSTRACT FROM AUTHOR]

Comprehensive, in-depth and accurate analyses of patent technology topic evolutions become increasingly significant since the analytical results can offer related personnel the scientific support to explore or trace back to the origin and the development of the technology. However, existing methods of topic evolutions do not facilitate better understanding of how a technology topic has evolved. This paper introduces an integrated method with the LDA topic identification analysis, the improved topic life cycle analysis, and the improved technology entropy analysis for identifying, measuring and interpreting topics evolutions from patent literatures. Multiple indicators we proposed and improved have been used to measure the degree of topic development and identify the topic types of different states. And, the concept of technology entropy has been redefined and improved to measure the changes of evolution intensity and evolution direction among topics, mainly used the topic word and its probability. The results from different methods are mutually connected and complemented. The process and characteristics of topic evolution are further overviewed. Graphene is selected for the case study. The mechanism of evolution and the effect of improved methods are focused on. The research has clearly shown that more accurate and comprehensive results can be achieved for topic evolution by employing this integrated method. Furthermore, the above integration of methods has potential contributions to hot spot detection and potential technology discovery. [ABSTRACT FROM AUTHOR]

Quantification of knowledge technologies has long posed a challenge to the measurement of cognitive proximity. This paper proposes a method to measure cognitive proximity by mining patent description text with the LDA topic model. With the patent-topic distribution got from the LDA topic model, the cognitive proximity is measured between enterprises or within cities, which could make up for the shortage of existing measurement methods limited by the rigid IPC, industry classification system, or non-standard interview data. Our empirical studies on the ICT industry indicate that the 20 topics obtained through the topic model have a good correspondence with the technologies involved in this industry's leading products and services. And we dig out the knowledge and technology information in the patent text to depict the technology landscape, including mining the changes of technology topics over time, the difference of distribution in various cities, and the development trend of the urban innovation network. This method's effectiveness is also proved in the model that compares different measurement methods when revealing the relationship between cognitive proximity and patent productivity. Last, researchers can use this approach to delve deeper into urban innovation issues, and policymakers can use it to figure out further innovation. [ABSTRACT FROM AUTHOR]

We propose two new topic modeling methods for sequential documents based on hybrid inter-document topic dependency. Topic modeling for sequential documents is the basis of many attractive applications such as emerging topic clustering and novel topic detection. For these tasks, most of the existing models introduce inter-document dependencies between topic distributions. However, in a real situation, adjacent emerging topics are often intertwined and mixed with outliers. These single-dependency based models have difficulties in handling the topic evolution in such multi-topic and outlier mixed sequential documents. To solve this problem, our first method considers three kinds of topic dependencies for each document to handle its probabilities of belonging to a fading topic, an emerging topic, or an independent topic. Secondly, we extend our first method by considering fine-grained dependencies in a given context for more complex topic evolution sequences. Our experiments conducted on six standard datasets on topic modeling show that our proposals outperform state-of-the-art models in terms of the accuracy of topic modeling, the quality of topic clustering, and the effectiveness of outlier detection. [ABSTRACT FROM AUTHOR]

Event-based social networks (EBSNs) facilitate people to interact with each other by sharing similar interests in online groups or taking part in offline events together. Event recommendation in EBSNs has been studied by many researchers. However, the problem of recommending the event to the top N active-friends of the key user has rarely been studied in EBSNs. In this paper, we propose a new method to solve this problem. In this method, we first construct an association matrix from the content of events and user features. Then, we define a new content-based event recommendation model, which combines the matrix, spatio-temporal relations and user interests to recommend an event to the active-friends of a key user. A series of experiments were conducted on real datasets collected from Meetup, and the comparison results have demonstrated the effectiveness of the new model. [ABSTRACT FROM AUTHOR]

Fine-grained entity typing (FGET) is to classify the mentions of entities into hierarchical fine-grained semantic types. There are two main issues with existing FGET approaches. Firstly, the process of training corpora for FGET is normally to label the data automatically, which inevitably induces noises. Existing approaches either directly tweak noisy labels in corpora by heuristics or algorithmically retreat to parental types, both leading to coarse-grained type labels instead of fine-grained ones. Secondly, existing approaches usually use recurrent neural networks to generate feature representations of mention phrases and their contexts, which, however, perform relatively poor on long contexts and out-of-vocabulary (OOV) words. In this paper, we propose a transfer learning-based approach to extract more efficient feature representations and offset label noises. More precisely, we adopt three transfer learning schemes: (i) transferring sub-word embeddings to generate more efficient OOV embeddings; (ii) using a pre-trained language model to generate more efficient context features; (iii) using a pre-trained topic model to transfer the topic-type relatedness through topic anchors and select confusing fine-grained types at inference time. The pre-trained topic model can offset the label noises without retreating to coarse-grained types. The experimental results demonstrate the effectiveness of our transfer learning approach for FGET. [ABSTRACT FROM AUTHOR]

The past few years have witnessed a substantial increase in functional rich API services and their compositions (e.g., Mashup services) on the Internet, which as a result proposes new requirement of organization and management methods for better service understanding, discovery and usage. Tagging is known to be efficient for this purpose, and many tag recommender systems for API services have been proposed in the past. Existing approaches targeted at API services usually recommend several similar tags, which is insufficient for Mashup services that normally demonstrate diverse functionalities. In this paper, we propose a novel approach for Mashup service tag completion that can recommend tags revealing the functional features of Mashup services more comprehensively. It first extracts candidate tags for the target Mashup from API services in different functional domains. Then, it adopts a hybrid filtering procedure to recommend the most relevant tags. To support efficient tag extraction and recommendation, an attention-based topic model called Att-LDA is proposed that can highlight the functional oriented features in descriptions of services for accurate functional semantic learning. Experiments and validations on a real-word dataset demonstrate the performance of our approach improved 7.1% compared with other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Social media platforms such as Twitter, Facebook, and YouTube have unique architecture, norms, and culture. These platforms are valuable sources of people's opinions which should be examined for knowledge discovery and user behavior analysis. This paper proposed a novel content analysis to examine user reviews or movie comments on YouTube. In fact, the proposed hybrid framework is based on semantic and sentiment aspects using fuzzy lattice reasoning to meaningful latent-topic detection and utilizing sentiment analysis of user comments of the Oscar-nominated movie trailers on YouTube. Based on the word vector feature, classification algorithms are employed to detect the comments' sentiment level. The results of this study suggest that the hybrid framework could be effective to extract features associated and latent topics with sentiment valence on user comments. In addition, NLP methods can have an impressive role for exploring the relationship between user opinion and Oscar movies comments on YouTube. [ABSTRACT FROM AUTHOR]

Identifying the evolution path of a research field is essential to scientific and technological innovation. There have been many attempts to identify the technology evolution path based on the topic model or social networks analysis, but many of them had deficiencies in methodology. First, many studies have only considered a single type of information (text or citation information) in scientific literature, which may lead to incomplete technology path mapping. Second, the number of topics in each period cannot be determined automatically, making dynamic topic tracking difficult. Third, data mining methods fail to be effectively combined with visual analysis, which will affect the efficiency and flexibility of mapping. In this study, we developed a method for mapping the technology evolution path using a novel non-parametric topic model, the citation involved Hierarchical Dirichlet Process (CIHDP), to achieve better topic detection and tracking of scientific literature. To better present and analyze the path, D3.js is used to visualize the splitting and fusion of the evolutionary path. We used this novel model to mapping the artificial intelligence research domain, through a successful mapping of the evolution path, the proposed method's validity and merits are shown. After incorporating the citation information, we found that the CIHDP can be mapping a complete path evolution process and had better performance than the Hierarchical Dirichlet Process and LDA. This method can be helpful for understanding and analyzing the development of technical topics. Moreover, it can be well used to map the science or technology of the innovation ecosystem. It may also arouse the interest of technology evolution path researchers or policymakers. [ABSTRACT FROM AUTHOR]

Citations are commonly used to measure academic impacts of scientific publications, including books. However, citation frequencies of books are single numerical evaluation metrics. It neglects details about books (e.g. contents), which may lead to the decline in comprehensiveness of evaluation results. Hence, fine-grained mining on books' citation information to integrate frequency metrics and content metrics can obtain more reliable evaluation results. Books' citation literatures (i.e. literatures cited books) present citation frequencies of books, and reflect citation intentions, topics and domains simultaneously. Existing research focused on analysing citation frequencies, authors or citation contexts of citation literatures to conduct citation analysis. It may be costly for collecting citation contexts and neglected latent information of citation literatures, such as impact scopes or topics of books reflected by citation literatures. Therefore, in this paper, we conducted fine-grained analysis on books' citation literatures to assess whether citation literatures could be systematically used for indicators of books' wider impacts. Specifically, we firstly collected books and corresponding information about their citation literatures. Then, we extracted multi-dimensional metrics via multi-granularity mining on citation literatures, and got assessment results by integrating content-level and frequency-level metrics. Finally, we compared assessment results based on citation literatures and existing metrics for assessing books' impacts to verify assessment results. Experimental results infer that citation literatures are a promising source for book impact assessment, especially books' academic impacts. [ABSTRACT FROM AUTHOR]

In recent years, with the popularity of mobile devices and mobile Internet, more and more social media sites are growing in an explosive way. Therefore, the social hot event will be rapidly fermented by the interaction of a large number of network users, and a large amount of multimedia data (such as texts, images and videos) will be generated. Therefore, it is important and necessary to conduct the research of multimedia social event analysis to know the evolutionary trend of social event over time automatically. This paper provides a survey and summarizes major progresses in multimedia social event analysis. We focus on four areas: (1) multimedia social event representation; (2) multimedia social event detection and tracking; (3) multimedia social event evolutionary analysis; and (4) multimedia social event topic mining. [ABSTRACT FROM AUTHOR]

In recent years, short text topic modeling has drawn considerable attentions from interdisciplinary researchers. Various customized topic models have been proposed to tackle the semantic sparseness nature of short texts. Most (if not all) of them follow the bag-of-words assumption, which, however, is not adequate since word order and phrases are often critical to capturing the meaning of texts. On the other hand, while some existing topic models are sensitive to word order, they do not perform well on short texts due to the severe data sparseness. To address these issues, we propose the Pseudo-document-based Topical N-Grams model (PTNG), which alleviates the data sparsity problem of short texts while is sensitive to word order. Extensive experiments on three real-world data sets with state-of-the-art baselines demonstrate the high quality of topics learned by PTNG according to UCI coherence scores and more discriminative semantic representation of short texts according to classification results. [ABSTRACT FROM AUTHOR]

As one of the fundamental information extraction methods, topic model has been widely used in text clustering, information recommendation and other text analysis tasks. Conventional topic models mainly utilize word co-occurrence information in texts for topic inference. However, it is usually hard to extract a group of words that are semantically coherent and have competent representation ability when the models applied into short texts. It is because the feature space of the short texts is too sparse to provide enough co-occurrence information for topic inference. The continuous development of word embeddings brings new representation of words and more effective measurement of word semantic similarity from concept perspective. In this study, we first mine word co-occurrence patterns (i.e., biterms) from short text corpus and then calculate biterm frequency and semantic similarity between its two words. The result shows that a biterm with higher frequency or semantic similarity usually has more similar words in the corpus. Based on the result, we develop a novel probabilistic topic model, named Noise Biterm Topic Model with Word Embeddings (NBTMWE). NBTMWE extends the Biterm Topic Model (BTM) by introducing a noise topic with prior knowledge of frequency and semantic similarity of biterm. NBTMWE shows the following advantages compared with BTM: (1) It can distinguish meaningful latent topics from a noise topic which consists of some common-used words that appear in many texts of the dataset; (2) It can promote a biterm's semantically related words to the same topic during the sampling process via generalized P ó l y a Urn (GPU) model. Using auxiliary word embeddings trained from a large scale of corpus, we report the results testing on two short text datasets (i.e., Sina Weibo and Web Snippets). Quantitatively, NBTMWE outperforms the state-of-the-art models in terms of coherence, topic word similarity and classification accuracy. Qualitatively, each of the topics generated by NBTMWE contains more semantically similar words and shows superior intelligibility. [ABSTRACT FROM AUTHOR]

Deceptive reviews on Web are a common phenomenon and how to detect them has a very important impact on products, services, and even business policies. In order to filter out deceptive reviews more accurately, a new model called Sentence Joint Topic Sentiment Model (SJTSM) is presented in this paper, which incorporates the sentence structure of reviews and the sentiment label information of words based on Latent Dirichlet Allocation (LDA) model to extract the review features. The proposed model employs Gibbs algorithm to estimate the maximum likelihood parameters and takes the vector of topic-sentiment distribution as the review features. Then a voting system of multiple-classifier, which takes the extracted review feature vector as its input is designed to realize the classification of deceptive review detection. The comparative experiments on different public datasets with other existing methods based on LDA model show that the new classifying system based on SJTSM model can achieve more satisfying classification results on deceptive review detection. [ABSTRACT FROM AUTHOR]

In common design pattern collections, e.g., design pattern books, design patterns are documented with templates that consist of multiple attributes, such as intent, structure, and sample code. To adapt to modern developers, the depictions of design patterns, especially some specific attributes, should advance with the current programming technologies, for example, "known uses", which exemplifies the use scenarios of design patterns in practice, and "related patterns", which describes the relatedness between a design pattern and the others within a context. However, it is not easy to update the contents of these attributes manually due to the diversity of the programming technologies. To address this problem, in this work, we conducted a case study to mine design pattern use scenarios and related design pattern pairs from Stack Overflow posts to enrich the two attributes. We first extracted the question posts relevant to each design pattern by identifying the design pattern tags. Then, the topics of the posts were discovered by applying topic modeling techniques. Finally, by analyzing the topics specified for each design pattern, we detected 195 design pattern use scenarios and 70 related design pattern pairs, involving 61 design patterns totally. These findings are associated with a variety of popular software frameworks and programming techniques. They could complement the existing design pattern collections and help developers better acknowledge the usage and relatedness of design patterns in today's programming practice. [ABSTRACT FROM AUTHOR]

Binary features, such as categories, keywords, or tags, are widely used to describe product properties. However, these features are incomplete in that they do not contain several aspects of numerical information. The qualitative score of tags is widely used to describe which product is better in terms of the given property. For example, in a restaurant navigation site, properties such as mood, dishes, and location are given in the form of numerical values, representing the goodness of each aspect. In this paper, we propose a novel approach to estimate the qualitative score from the binary features of products. Based on a natural assumption that an item with a better property is more popular among users who prefer that property, in short, "experts know best," we introduce both discriminative and generative models with which user preferences and item qualitative scores are inferred from user--item interactions. We constrain the space of the item qualitative score by item binary features so that the score of each item and tag can only have nonzero values when the item has the corresponding tag. This approach contributes to resolving the following difficulties: (1) no supervised data for the score estimation, (2) implicit user purpose, and (3) irrelevant tag contamination. We evaluate our models by using two artificial datasets and two real-world datasets of movie and book ratings. In the experiment, we evaluate the performances of our model under sparse transaction and noisy tag settings by using two artificial datasets. We also evaluate our models' resolution for irrelevant tags using the real-world dataset of movie ratings and observe that our models outperform a baseline model. Finally, tag rankings obtained from the real-world datasets are compared with a baseline model. [ABSTRACT FROM AUTHOR]

Topic models have been widely used to infer latent topics in text documents. However, the unsupervised topic models often result in incoherent topics, which always confused users in applications. Incorporating prior domain knowledge into topic models is an effective strategy to extract coherent and meaningful topics. In this paper, we go one step further to explore how different forms of prior semantic relations of words can be encoded into models to improve the performance of topic modeling process. We develop a novel topic model—called Mixed Word Correlation Knowledge-based Latent Dirichlet Allocation—to infer latent topics from text corpus. Specifically, the proposed model mines two forms of lexical semantic knowledge based on recent progress in word embedding, which can represent semantic information of words in a continuous vector space. To incorporate generated prior knowledge, a Mixed Markov Random Field is constructed over the latent topic layer to regularize the topic assignment of each word during the topic sampling process. Experimental results on two public benchmark datasets illustrate the superior performance of the proposed approach over several state-of-the-art baseline models. [ABSTRACT FROM AUTHOR]

In the past decades, there have been a number of proposals to apply topic modeling to research trend analysis. However, most of previous studies have relied primarily on document publication year and have not incorporated the impact of articles into trend analysis. Unlike previous trend analysis using topic modeling, we incorporate citation count, which can be viewed as the impact of articles, into trend analysis to shed a new light on the understanding of research trends. To this end, we propose the Generalized Dirichlet multinomial regression (g-DMR) topic model, which improves the DMR topic model by replacing a linear inner product in topic priors, exp x d · λ t , with a more general form based on topic distribution function (TDF), exp f x d + ε . We use multidimensional Legendre Polynomial as TDF to capture publication year and the number of citations per publication simultaneously. In DMR model, since metadata could affect the document-topic distribution only monotonically and continuous values such as publication year and citation count need to be discretized, it is difficult to view the dynamic change of each topic. But the g-DMR model can handle various orthogonal continuous variables with arbitrary order of polynomial, so it can show more dynamic topic trends. Two major experiments show that the proposed model is better suited for topic generation with consideration of citation impact than DMR does for the trend analysis in the field of Library and Information Science in general and Text Mining in particular. [ABSTRACT FROM AUTHOR]

The objective of description models is to generate image captions to elaborate contents. Despite recent advancements in machine learning and computer vision, generating discriminative captions still remains a challenging problem. Traditional approaches imitate frequent language patterns without considering the semantic alignment of words. In this work, an image captioning framework is proposed that generates topic sensitive descriptions. The model captures the semantic relation and polysemous nature of the words that describe the images and resultantly generates superior descriptions for the target images. The efficacy of the proposed model is indicated by the evaluation on the state-of-the-art captioning datasets. The model shows promising performance compared to the existing description models proposed in the recent literature. [ABSTRACT FROM AUTHOR]

Scientific networks play an increasingly important role in facilitating knowledge and technique diffusion. In such networks, highly influential nodes (scientists or literatures) are prone to stimulate other researchers in the generation of innovative ideas. The objective of this study is to detect topic-level influencers from a large collection of links between nodes and textual contents in scientific networks. For this purpose, we propose a sparse link topic model (SLTM) that introduces a "Spike and Slab" prior to achieve sparsity in node-topic distribution. Compared with previous approaches, our model assumes that a node usually focuses on several salient topics instead of a wide range of topics, which is useful in learning topic-level influencers in scientific networks. In addition, a collapsed variational Bayesian (CVB) inference algorithm is designed for large-scale applications. Our experiments are conducted on a large scientific collaboration network. The results reveal that the proposed model significantly improves the precision of topic-level detection. Our analysis also reflects that SLTM can explicitly model the sparse topical structure of each node in the network. [ABSTRACT FROM AUTHOR]