Search

Service quality is a multi-dimensional construct which is not accurately measured by aspects deriving from numerical ratings and their associated weights. Extant literature in the expert and intelligent systems examines this issue by relying mainly on such constrained information sets. In this study, we utilize online reviews to show the information gains from the consideration of factors identified from topic modeling of unstructured data which provide a flexible extension to numerical scores to understand customer satisfaction and subsequently service quality. When numerical and textual features are combined, the explained variation in overall satisfaction improves significantly. We further present how such information can be of value for firms for corporate strategy decision-making when incorporated in an expert system that acts as a tool to perform market analysis and assess their competitive performance. We apply our methodology on airline passengers’ online reviews using Structural Topic Models (STM), a recent probabilistic extension to Latent Dirichlet Allocation (LDA) that allows the incorporation of document level covariates. This innovation allows us to capture dominant drivers of satisfaction along with their dynamics and interdependencies. Results unveil the orthogonality of the low-cost aspect of airline competition when all other service quality dimensions are considered, thus explaining the success of low-cost carriers in the airline market.

As the rapid development of Web 2.0 communities, social media service providers offer users a convenient way to share and create their own contents such as online comments, blogs, microblogs/tweets, etc. Understanding the latent emotions of such short texts from social media via the computational model is an important issue as such a model will help us to identify the social events and make better decisions (e.g., investment in stocking market). However, it is always very challenge to detect emotions from above user-generated contents due to the sparsity problem (e.g., a tweet is a short message). In this article, we propose an universal affective model (UAM) to classify readers’ emotions over unlabeled short texts. Different from conventional text classification model, the UAM structurally consists of topic-level and term-level sub-models, and detects social emotions from the perspective of readers in social media. Through the evaluation on real-world data sets, the experimental results validate the effectiveness of the proposed model in terms of the effectiveness and accuracy.

Event detection over microblogs has attracted great research interest due to its wide application in crisis management and decision making etc. In natural disasters, complex events are reported in real time on social media sites, but these reports are invisible to crisis coordinators. Detecting these crisis events helps watchers to make right decisions rapidly, reducing injuries, deaths and economic loss. In sporting activities, detecting events helps audiences make better and more timely game viewing plans. However, existing event detection techniques are not effective at handling complex social events that evolve over time. In this paper, we propose an event detection method that takes advantage of retweeting behavior for handling the events evolution. Specifically, we first propose a topic model called RL-LDA to capture the social media information over hashtag, location, textual and retweeting behavior. Using RL-LDA, a complex event can be well handled by exploring the correlation between retweeting behavior and the event. Then to maintain the RL-LDA in a dynamic environment, we propose a dynamic update algorithm, which incrementally updates events over real time streams. Experiments over real-world datasets show that RL-LDA detects the temporal evolution of complex events effectively and efficiently.

Traditional topic modeling has been widely studied and popularly employed in expert systems and information systems. However, traditional topic models cannot discover structural relations among topics, thus losing the chance to explore the data more deeply. Hierarchical topic modeling has the capability of learning topics, as well as discovering the hierarchical topic structure from text data. But purely unsupervised models tend to generate weak topic hierarchies. To solve this problem, we propose a novel knowledge-based hierarchical topic model (KHTM), which can incorporate prior knowledge into topic hierarchy building. A key novelty of this model is that it can mine prior knowledge automatically from the topic hierarchies of multiple domains corpora. In this paper, the knowledge is represented as the word pairs which satisfy the requirement of frequent co-occurrence, and knowledge is organized in form of hierarchical structure. We also propose an iterative learning algorithm. For evaluation, we crawled two new multi-domain datasets and conducted comprehensive experiments. The experimental results show that our algorithm and model can generate more coherent topics, and more reasonable hierarchical structure.

Social image annotation, which aims at inferring a set of semantic concepts for a social image, is an effective and straightforward way to facilitate social image search. Conventional approaches mainly demonstrated on adopting the visual features and tags, without considering other types of metadata. How to enhance the accuracy of social image annotation by fully exploiting multi-modal features is still an opening and challenging problem. In this paper, we propose an improved Multi-Modal Data Fusion based Latent Dirichlet Allocation (LDA) topic model (MMDF-LDA) to annotate social images via fusing visual content, user-supplied tags, user comments, and geographic information. When MMDF-LDA samples annotations for one data modality, all the other data modalities are exploited. In MMDF-LDA, geographical topics are generated from GPS locations of social images, and annotations have different probability to be used in different geographical regions. A social image is divided into several patches in advance, and then MMDF-LDA assigns annotations for the patches of social images by estimating the probability of annotation-patch assignment. Through experiments in social image annotation and retrieval on several datasets, we demonstrate the effectiveness of the proposed MMDF-LDA model in comparison with state-of-the-art methods.

With the increase of online customer opinions in specialised websites and social networks, automatic systems to help organise and classify customer reviews by domain-specific aspect categories and sentiment polarity are more needed than ever. Supervised approaches for Aspect Based Sentiment Analysis achieve good results for the domain and language they are trained on, but manually labelling data to train supervised systems for all domains and languages is very costly and time consuming. In this work, we describe W2VLDA, an almost unsupervised system based on topic modelling that, combined with some other unsupervised methods and a minimal configuration step, performs aspect category classification, aspect-term and opinion-word separation and sentiment polarity classification for any given domain and language. We evaluate its domain aspect and sentiment classification performance in the multilingual SemEval 2016 task 5 (ABSA) dataset. We show competitive results for several domains (hotels, restaurants, electronic devices) and languages (English, Spanish, French and Dutch).

Recommender systems are challenging research problems being exploited to suggest new items or services, such as books, music and movies, and even people, to users based on information about the user profile or the recommended items. To date, collaborative filtering (CF) has become one of the most widely used approaches for recommendations. However, traditional CF methods usually cannot track temporal dynamic user preferences and topic changes to make appropriate suggestions. Moreover, the performance of CF is limited in the case of sparse data. In this paper, we propose a novel temporal recommender system based on multiple transitions in user preference drift, called MTUPD, which employs a multitransition factor and a forgetting time function to investigate the evolution of user preferences. In addition, we consider addressing the rating sparsity issue by using text reviews. Understanding the reviews can facilitate the system grasping whether or not a user is attracted by the appearance of an item and whether the facet of an item’s appearance contributes the most to its ratings. To achieve this, we apply a topic model that automatically classifies hidden topic factors in each time period and incorporate the transition method for both user preferences and relevant review topics. Experiments show that our proposed model outperforms the compared models on eight promising datasets for temporal recommender systems.

A topic model based approach for novel summarization is proposed.An importance evaluation function for sentence candidates is designed.A summary smoothing approach is presented to improve the summary readability. Most of existing text automatic summarization algorithms are targeted for multi-documents of relatively short length, thus difficult to be applied immediately to novel documents of structure freedom and long length. In this paper, aiming at novel documents, we propose a topic modeling based approach to extractive automatic summarization, so as to achieve a good balance among compression ratio, summarization quality and machine readability. First, based on topic modeling, we extract the candidate sentences associated with topic words from a preprocessed novel document. Second, with the goals of compression ratio and topic diversity, we design an importance evaluation function to select the most important sentences from the candidate sentences and thus generate an initial novel summary. Finally, we smooth the initial summary to overcome the semantic confusion caused by ambiguous or synonymous words, so as to improve the summary readability. We evaluate experimentally our proposed approach on a real novel dataset. The experiment results show that compared to those from other candidate algorithms, each automatic summary generated by our approach has not only a higher compression ratio, but also better summarization quality.

A patent gives the owner of an invention the exclusive rights to make, use and sell their invention. Before a new patent application is filed, patent lawyers are required to engage in Prior Art Search to determine the likelihood that an invention is novel, valid or to make sense of the domain. To perform this search, existing platforms utilize keywords and Boolean Logic, which disregards the syntax and semantics of natural language and thus, making the search extremely difficult. Consequently, studies regarding semantics using neural embeddings exist, but these only consider a narrow number of unidirectional words. In this study, we propose an end-to-end framework to consider bidirectional semantics, syntax and the thematic nature of natural language for prior art search. The proposed framework goes beyond keywords as input queries and takes a patent as the input. The contributions of this paper is twofold; adapting pre-trained embedding models (e.g., BERT) to address the semantics and syntax of language, followed by the second component, which exploits topic modeling to build a diversified answer that covers all themes across domains of the input patent. We evaluate the performance of the proposed framework on the CLEF-IP 2011 benchmark dataset and a real-world dataset obtained from Google patent repository and show that the proposed framework outperforms existing methods and returns meaningful results for a given patent.

Text has traditionally been used to train automated classifiers for a multitude of purposes, such as: classification, topic modelling and sentiment analysis. State-of-the-art LSTM classifier require a large number of training examples to avoid biases and successfully generalise. Labelled data greatly improves classification results, but not all modern datasets include large numbers of labelled examples. Labelling is a complex task that can be expensive, time-consuming, and potentially introduces biases. Data augmentation methods create synthetic data based on existing labelled examples, with the goal of improving classification results. These methods have been successfully used in image classification tasks and recent research has extended them to text classification. We propose a method that uses sentence permutations to augment an initial dataset, while retaining key statistical properties of the dataset. We evaluate our method with eight different datasets and a baseline Deep Learning process. This permutation method significantly improves classification accuracy by an average of 4.1%. We also propose two more text augmentations that reverse the classification of each augmented example, antonym and negation. We test these two augmentations in three eligible datasets, and the results suggest an -averaged, across all datasets-improvement in classification accuracy of 0.35% for antonym and 0.4% for negation, when compared to our proposed permutation augmentation.

Humans must easily handle the vast amounts of data being generated by the revolution of information technology. Thus, Automatic Text summarization has been applied to various domains in order to find the most relevant information and make critical decisions quickly. In the context of Arabic, text summarization techniques suffer from several problems. First, most existing methods do not consider the context or domain to which the document belongs. Second, the majority of the existing approaches are based on the traditional bag-of-words representation, which involves high dimensional and sparse data, and makes it difficult to capture relevant information. Third, research in Arabic Text summarization is fairly small and only recently compared to that on Anglo-Saxon and other languages due to the shortage of Arabic corpora, resources, and automatic processing tools. In this paper, we try to overcome these limitations by proposing a new approach using documents clustering, topic modeling, and unsupervised neural networks in order to build an efficient document representation model. First, a new document clustering technique using Extreme learning machine is performed on large text collection. Second, topic modeling is applied to documents collection in order to identify topics present in each cluster. Third, each document is represented in a topic space by a matrix where rows represent the document sentences and columns represent the cluster topics. The generated matrix is then trained using several unsupervised neural networks and ensemble learning algorithms in order to build an abstract representation of the document in the concept space. Important sentences are ranked and extracted according to a graph model with a redundancy elimination component. The proposed approach is evaluated on Essex Arabic Summaries Corpus and compared against other Arabic text summarization approaches using ROUGE measure. Experimental results showed that the models trained on topic representation learn better representations and improve significantly the summarization performance. In particular, ensemble learning models demonstrated an important improvement on Rouge recall and promising results on F-measure.

Inspired by the importance of social media, a Social Network Opinion Leaders (SNOL) system has been proposed in this paper. The purpose of this system is to identify topic-based opinion leaders of social media. In order to accomplish this goal, several steps have been taken, such as data collection, data processing, data analysis, data classification, ranking of topic-based opinion leaders, and evaluation. The SNOL system has two main parts. In the first part, collected tweets are classified by semantic kernels for topic-based analysis. In the second part, leadership scores are given to each user in the network according to topic modeling and user modeling results. Leadership scores are then calculated with the formula generated and opinion leaders are determined for each category. Experiments are performed on data gathered from Twitter including 17,234,924 tweets from 38,727 users. The evaluation of opinion leader detection is a difficult job since there is no standard method for identifying opinion leaders. Therefore, the evaluation of the results of this study has been done using two different methods, retweet count and spread score, to prove that the suggested methodology outperforms the PageRank algorithm. The results have also been evaluated considering the user-topic sentiment correlation of the retrieved lists. Furthermore, SNOL has been compared against some opinion leader detection methods previously presented in the literature. The experimental results show that SNOL generates remarkably higher performance than the PageRank algorithm and other existing algorithms in the literature for nearly all topics and all selected top N opinion leaders.

The purpose of this paper is to develop a technology-based model for identifying various criteria in a decision-making situation. We used topic modeling to discover critical criteria and their corresponding weights in the Analytic Hierarchy Process (AHP). Approximately 100,000 hotel reviews and 100,000 restaurant reviews were scraped from TripAdvisor.com for criteria determination. Next, an AHP model with criteria and 12 hotels/restaurants as alternatives were compared and ranked. The results compared favorably with more than 1000 reviews of these hotels/restaurants in TripAdvisor.com, thus validating the methodology.

© 2020 Elsevier LtdWith the widespread use of social networks, blogs, forums and e-commerce web sites, the volume of user generated textual data is growing exponentially. User opinions in product reviews or in other textual data are crucial for manufacturers, retailers and providers of the products and services. Therefore, sentiment analysis and opinion mining have become important research areas. In user reviews mining, topic modeling based approaches and Latent Dirichlet Allocation (LDA) are significant methods that are used in extracting product aspects in aspect based sentiment analysis. However, LDA cannot be directly applied on user reviews and on other short texts because of data sparsity problem and lack of co-occurrence patterns. Several studies have been published for the adaptation of LDA for short texts. In this study, a novel method for aspect based sentiment analysis, Sentence Segment LDA (SS-LDA) is proposed. SS-LDA is a novel adaptation of LDA algorithm for product aspect extraction. The experimental results reveal that SS-LDA is quite competitive in extracting products aspects.

Method to extract aspects by incorporating domain knowledge into the LDA model.The knowledge is automatically generated based on co-occurrence relations.Knowledge validation is used to prevent incorrect and general knowledge.The method is language independent, allowing it to be applied to all languages. Aspect extraction is one of the fundamental steps in analyzing the characteristics of opinions, feelings and emotions expressed in textual data provided for a certain topic. Current aspect extraction techniques are mostly based on topic models; however, employing only topic models causes incoherent aspects to be generated. Therefore, this paper aims to discover more precise aspects by incorporating co-occurrence relations as prior domain knowledge into the Latent Dirichlet Allocation (LDA) topic model. In the proposed method, first, the preliminary aspects are generated based on LDA. Then, in an iterative manner, the prior knowledge is extracted automatically from co-occurrence relations and similar aspects of relevant topics. Finally, the extracted knowledge is incorporated into the LDA model. The iterations improve the quality of the extracted aspects.The competence of the proposed ELDA for the aspect extraction task is evaluated through experiments on two datasets in the English and Persian languages. The experimental results indicate that ELDA not only outperforms the state-of-the-art alternatives in terms of topic coherence and precision, but also has no particular dependency on the written language and can be applied to all languages with reasonable accuracy. Thus, ELDA can impact natural language processing applications, particularly in languages with limited linguistic resources.

A novel text classification method for learning from very small labeled set.The method uses a text representation based on the LDA topic model.Self-training is used to enlarge labeled set from unlabeled instances.A model for setting methods parameters for any document collection is proposed. Supervised text classification methods are efficient when they can learn with reasonably sized labeled sets. On the other hand, when only a small set of labeled documents is available, semi-supervised methods become more appropriate. These methods are based on comparing distributions between labeled and unlabeled instances, therefore it is important to focus on the representation and its discrimination abilities. In this paper we present the ST LDA method for text classification in a semi-supervised manner with representations based on topic models. The proposed method comprises a semi-supervised text classification algorithm based on self-training and a model, which determines parameter settings for any new document collection. Self-training is used to enlarge the small initial labeled set with the help of information from unlabeled data. We investigate how topic-based representation affects prediction accuracy by performing NBMN and SVM classification algorithms on an enlarged labeled set and then compare the results with the same method on a typical TF-IDF representation. We also compare ST LDA with supervised classification methods and other well-known semi-supervised methods. Experiments were conducted on 11 very small initial labeled sets sampled from six publicly available document collections. The results show that our ST LDA method, when used in combination with NBMN, performed significantly better in terms of classification accuracy than other comparable methods and variations. In this manner, the ST LDA method proved to be a competitive classification method for different text collections when only a small set of labeled instances is available. As such, the proposed ST LDA method may well help to improve text classification tasks, which are essential in many advanced expert and intelligent systems, especially in the case of a scarcity of labeled texts.

A new framework to retrieve semantically relevant images from the social database.Probabilistic topic model to predict the missing tags and remove the noisy ones.Two algorithms for the estimation of model parameters and tag correspondence.The scoring scheme relies on the fusion of visual and textual information.The outperformance of image annotation and retrieval to state-of-the-art methods. Nowadays, due to the rapid growth of digital technologies, huge volumes of image data are created and shared on social media sites. User-provided tags attached to each social image are widely recognized as a bridge to fill the semantic gap between low-level image features and high-level concepts. Hence, a combination of images along with their corresponding tags is useful for intelligent retrieval systems, those are designed to gain high-level understanding from images and facilitate semantic search. However, user-provided tags in practice are usually incomplete and noisy, which may degrade the retrieval performance. To tackle this problem, we present a novel retrieval framework that automatically associates the visual content with textual tags and enables effective image search. To this end, we first propose a probabilistic topic model learned on social images to discover latent topics from the co-occurrence of tags and image features. Moreover, our topic model is built by exploiting the expert knowledge about the correlation between tags with visual contents and the relationship among image features that is formulated in terms of spatial location and color distribution. The discovered topics then help to predict missing tags of an unseen image as well as the ones partially labeled in the database. These predicted tags can greatly facilitate the reliable measure of semantic similarity between the query and database images. Therefore, we further present a scoring scheme to estimate the similarity by fusing textual tags and visual representation. Extensive experiments conducted on three benchmark datasets show that our topic model provides the accurate annotation against the noise and incompleteness of tags. Using our generalized scoring scheme, which is particularly advantageous to many types of queries, the proposed approach also outperforms state-of-the-art approaches in terms of retrieval accuracy.

We combine topic model with Wikipedia knowledge.We represent a document as a bag of Wikipedia articles.Using Wikipedia page view statistics.Extracting more accurate dynamic patterns with specific and coherent entities.Spending less time by using Wikipedia knowledge. Probabilistic topic models could be used to extract low-dimension aspects from document collections, and capture how the aspects change over time. However, such models without any human knowledge often produce aspects that are not interpretable. In recent years, a number of knowledge-based topic models and dynamic topic models have been proposed, but they could not process concept knowledge and temporal information in Wikipedia. In this paper, we fill this gap by proposing a new probabilistic modeling framework which combines both data-driven topic model and Wikipedia knowledge. With the supervision of Wikipedia knowledge, we could grasp more coherent aspects, namely, concepts, and detect the trends of concepts more accurately, the detected concept trends can reflect bursty content in text and people's concern. Our method could detect events and discover events specific entities in text. Experiments on New York Times and TechCrunch datasets show that our framework outperforms two baselines.

Electronic commerce has become a popular shopping mode. To enhance their reputations, attract more customers, and finally obtain more benefits, dishonest sellers often recruit buyers or robots to post a large number of deceptive reviews to mislead users. According to the interpretability of learning results, existing methods for detecting deceptive reviews can be mainly divided into explicit feature-based mining ones and neural network-based implicit feature mining ones. The nature of these works is accurate text classification based on coarse-grained features (e.g., topic, sentence, and document) or fine-grained features (e.g., word). To take full merits of existing approaches, this paper proposes a new framework that explores a method to combine the coarse-grained features and the fine-grained features. In this framework, the coarse-grained implicit semantic features of the topic distribution are learned by the concatenation of a Latent Dirichlet Allocation (LDA) topic model and a 2-layered neural network. The fine-grained implicit semantic features from the word vectors representation of the reviews are parallelly learned by a deep learning framework. Finally, these two granular features are combined and adopted to train a Support Vector Machine (SVM) classifier for detecting whether a review is deceptive or not. To verify the effectiveness and performance of this framework, we derive three models by specifying three popular deep learning models, such as TextCNN, long short-term memory (LSTM), and Bi-directional LSTM (BiLSTM) to learn the fine-grained features. Experimental results on a mixed-domain dataset and balanced/unbalanced in-domain datasets show that all the combination models are superior to the corresponding baseline models considering single features.

We propose a learning framework for predictive monitoring of clinical pathways.Both offline analysis and online monitoring services are provided.A probabilistic topic model is developed to describe essential behaviors of CPs.Two predictive monitoring services are presented to illustrate the potential of our framework.Evaluation is done using a real clinical datasetfrom a Chinese hospital. ObjectiveAccurate and timely monitoring, as a key aspect of clinical pathway management, provides crucial information to medical staff and hospital managers for determining the efficient medical service delivered to individual patients, and for promptly handling unusual treatment behaviors in clinical pathways (CPs). In many applications, CP monitoring is performed in a reactive manner, e.g., variant treatment events are detected only after they have occurred in CPs. Alternatively, this article presents an intelligent learning system for predictive monitoring of CPs and from a large volume of electronic medical records (EMRs). MethodsThe proposed system is composed of both offline analysis and online monitoring phases. In the offline phase, a particular probabilistic topic model, i.e., treatment pattern model (TPM), is generated from electronic medical records to describe essential/critical medical behaviors of CPs. Using TPM-based measures as a descriptive vocabulary, online monitoring of CPs can be provided for ongoing patient-care journeys. Specifically, this article presents two typical predictive monitoring services, i.e., unusual treatment event prediction and clinical outcome prediction, to illustrate how the potential of the proposed system can be exploited to provide online monitoring services from both internal and external perspectives of CPs. ResultsThe proposed monitoring services have been evaluated using a real clinical dataset pertaining to the unstable angina CP and collected from a large hospital in China. In terms of unusual treatment event prediction, the overall precision and recall of our system are 0.834, and 0.96, respectively, which is comparable to identify unusual treatment events in CPs in comparison with human evaluation. In terms of clinical outcome prediction, the stable model was characterized by 0.849 accuracy, 0.064 hamming-loss and 0.053 one-loss, which outperforms the benchmark multi-label classification algorithms on clinical outcome prediction. ConclusionExtensive evaluations on a real clinical data-set, typically missing from other work, demonstrate that the proposed system, as a crucial advantage over traditional expert systems for CP management, not only provides an efficient and general surveillance of CPs, but also empowers clinicians with the capability to look insights into CPs to gain a deeper understanding of the situations in which the proposed prediction technique performs well.

A novel method for predicting stock price movement was presented.Topics and sentiments of them were extracted from social media as the feature.Two methods were proposed to capture the topic-sentiment feature.Integration of the sentiments was investigated via a large scale experiment.Our model outperformed other methods in the average accuracy of 18 stocks. The goal of this research is to build a model to predict stock price movement using the sentiment from social media. Unlike previous approaches where the overall moods or sentiments are considered, the sentiments of the specific topics of the company are incorporated into the stock prediction model. Topics and related sentiments are automatically extracted from the texts in a message board by using our proposed method as well as existing topic models. In addition, this paper shows an evaluation of the effectiveness of the sentiment analysis in the stock prediction task via a large scale experiment. Comparing the accuracy average over 18 stocks in one year transaction, our method achieved 2.07% better performance than the model using historical prices only. Furthermore, when comparing the methods only for the stocks that are difficult to predict, our method achieved 9.83% better accuracy than historical price method, and 3.03% better than human sentiment method.

Blockchain has become one of the core technologies in Industry 4.0. To help decision-makers establish action plans based on blockchain, it is an urgent task to analyze trends in blockchain technology. However, most of existing studies on blockchain trend analysis are based on effort demanding full-text investigation or traditional bibliometric methods whose study scope is limited to a frequency-based statistical analysis. Therefore, in this paper, we propose a new topic modeling method called Word2vec-based Latent Semantic Analysis (W2V-LSA), which is based on Word2vec and Spherical k-means clustering to better capture and represent the context of a corpus. We then used W2V-LSA to perform an annual trend analysis of blockchain research by country and time for 231 abstracts of blockchain-related papers published over the past five years. The performance of the proposed algorithm was compared to Probabilistic LSA, one of the common topic modeling techniques. The experimental results confirmed the usefulness of W2V-LSA in terms of the accuracy and diversity of topics by quantitative and qualitative evaluation. The proposed method can be a competitive alternative for better topic modeling to provide direction for future research in technology trend analysis and it is applicable to various expert systems related to text mining.

Openness to experience, one of the essential individual characteristics, is of great theoretical and practical value in psychological and behavioral domains. Although typical machine learning methods can be utilized to extract individuals’ openness to experience from the large-scale textual data like the unprecedented massive user generated contents (UGCs), they are often regarded as “black boxes” because they are unable to provide knowledge about the influential factors of openness to experience. This is of no help for us to investigate why a particular level of openness to experience is predicted for an individual. In addition, high dimensionality and sparseness of textual data impairs the performance of the typical machine learning method in extracting individuals’ characteristics. In this study, we propose an interpretable data-driven mixture method for qualified modeling and predicting individuals’ openness to experience. The proposed method extends the latent Dirichlet allocation (LDA) to overcome the problem of high dimensionality and sparseness in modeling the textual data, and can effectively extract two influential variables, namely, the topic preference and the expressed emotional intensity, to make an accurate prediction and to help us fully understand individuals’ openness to experience lurking in the textual data. Experimental results indicate the effectiveness of the proposed method in drawing individuals’ openness to experience, and also validate the predictive ability of topic preference and expressed emotional intensity which are indicated in psychological literature to be influential factors of openness to experience.

A reading comprehension tool for people with ASD is exposed.The tool assist people with ASD and careers to simplify texts.A key element of the tool is the concepts substitution with images.Figurative language is also simplified for people with ASD.Summary generation and topic detection are other functions of the tool. People affected by Autism Spectrum Disorders (ASD) have impairments in social interaction because they lack an adequate theory of mind. A significant percentile has inadequate reading comprehension skills. We present a multilingual tool called Open Book (OB) that applies Human Language Technologies (HLT) in order to identify reading comprehension obstacles in text documents and propose more simple alternatives with the aim of assisting the reading comprehension of users. OB involves several text transformations at lexical, syntactic and semantic level. In this paper we focus on three challenging components of the OB tool: the image retrieval component, the idiom detection component and the summarization module. There are very few studies that involve simplification by showing images associated to difficult concepts. In addition, the treatment of figurative language such as idioms or metaphors is one of the most challenging areas in Natural Language Processing (NLP). Finally, although text summarization is a more widely studied field in NLP, its application to text simplification remains as an open research issue. Thus, we focus on the integration of these three modules in our OB tool. We present the motivation for building these components and we describe how they are integrated in the whole system. Moreover, the usability and the usefulness of OB have been evaluated and analysed showing that the tool helps to produce texts easier to understand for autistic people.

A novel contextual topic model is proposed for multi-document summarization.The main idea is to leverage hierarchical topics and their correlations with respect to the lexical co-occurrences of words.The proposed contextual topic model can effectively determine the relevance of sentences. Information overload becomes a serious problem in the digital age. It negatively impacts understanding of useful information. How to alleviate this problem is the main concern of research on natural language processing, especially multi-document summarization. With the aim of seeking a new method to help justify the importance of similar sentences in multi-document summarizations, this study proposes a novel approach based on recent hierarchical Bayesian topic models. The proposed model incorporates the concepts of n-grams into hierarchically latent topics to capture the word dependencies that appear in the local context of a word. The quantitative and qualitative evaluation results show that this model has outperformed both hLDA and LDA in document modeling. In addition, the experimental results in practice demonstrate that our summarization system implementing this model can significantly improve the performance and make it comparable to the state-of-the-art summarization systems.

An efficient framework to classify short text from scientific documents is proposed.Topic models from various types of knowledge were used for enhancing features in documents.Two methods were presented to optimize external features that enhance relatedness in documents.The performances were evaluated by using real-world scientific documents from online publisher.Proposed methods are shown to outperform related work. Classification of short text is challenging due to data sparseness, which is a typical characteristic of short text. In this paper, we propose methods for enhancing features using topic models, which make short text seem less sparse and more topic-oriented for classification. We exploited topic model analysis based on Latent Dirichlet Allocation for enriched datasets, and then we presented new methods for enhancing features by combining external texts from topic models that make documents more effective for classification. In experiments, we utilized the title contents of scientific articles as short text documents, and then enriched these documents using topic models from various types of universal datasets for classification in order to show that our approach performs efficiently.

We present an approach for the dynamic adaptation of the LMs used by a speech recognizer based on topic identification.For each audio segment the system interpolates a static LM with a topic-based LM in a two stages recognition architecture.The interpolation weights are computed based on different sources of information (distance of LMs and topic similarity).We evaluate different strategies for generating topic-specific LMs and for the adaptation of the LM used in the final stage.Evaluation shows a significant reduction of the error rates for both tasks (topic identification and speech recognition). In this paper we present an efficient speech recognition approach for multitopic speech by combining information retrieval techniques and topic-based language modeling. Information retrieval based techniques, such as topic identification by means of Latent Semantic Analysis, are used to identify the topic in a recognized transcription of an audio segment. According to the confidence on the topics that have been identified, we propose a dynamic language model adaptation in order to improve the recognition performance in 'a two stages' automatic speech recognition system. The scheme used for the adaptation of the language model is a linear interpolation between a background general LM and a topic dependent LM. We have studied different approaches to generate the topic dependent LM and also for determining the interpolation weight of this model with the background model. In one of these approaches we use the given topic labels in the training dataset to obtain the topic models. In the other approach we separate the documents in the training dataset into topic clusters by using the k-means algorithm. For strengthening the adaptation models we also use topic identification techniques to group non topic-labeled documents from the EUROPARL text database in order to increase the amount of data for training specific topic based language models. For the evaluation of the proposed system we are using the Spanish partition of the European Parliament Plenary Sessions (EPPS) Database; we selected a subset of the database with 67 labeled topics for the evaluation. For the task of topic identification our experiments show a relative reduction in topic identification error of 44.94% when compared to the baseline method, the Generalized Vector Model with a classic TF-IDF weighting scheme. For the task of dynamic adaptation of LMs applied to ASR we have achieved a relative reduction in WER of 13.52% over a single background language model.

Social media platforms such as Twitter are becoming increasingly mainstream which provides valuable user-generated information by publishing and sharing contents. Identifying interesting and useful contents from large text-streams is a crucial issue in social media because many users struggle with information overload. Retweeting as a forwarding function plays an important role in information propagation where the retweet counts simply reflect a tweet's popularity. However, the main reason for retweets may be limited to personal interests and satisfactions. In this paper, we use a topic identification as a proxy to understand a large number of tweets and to score the interestingness of an individual tweet based on its latent topics. Our assumption is that fascinating topics generate contents that may be of potential interest to a wide audience. We propose a novel topic model called Trend Sensitive-Latent Dirichlet Allocation (TS-LDA) that can efficiently extract latent topics from contents by modeling temporal trends on Twitter over time. The experimental results on real world data from Twitter demonstrate that our proposed method outperforms several other baseline methods.

The tourism industry has experienced a shift from offline to online travellers and this has made the use of intelligent systems in the tourism sector crucial. These information systems should provide tourism consumers and service providers with the most relevant information, more decision support, greater mobility and the most enjoyable travel experiences. As a consequence, Destination Marketing Organizations (DMOs) not only have to respond by adopting new technologies, but also by interpreting and using the knowledge created by the use of these techniques. This work presents the design of a general and non-invasive web mining system, built using the minimum information stored in a web server (the content of the website and the information from the log files stored in Common Log Format (CLF)) and its application to the Bidasoa Turismo (BTw) website. The proposed system combines web usage and content mining techniques with the three following main objectives: generating user navigation profiles to be used for link prediction; enriching the profiles with semantic information to diversify them, which provides the DMO with a tool to introduce links that will match the users taste; and moreover, obtaining global and language-dependent user interest profiles, which provides the DMO staff with important information for future web designs, and allows them to design future marketing campaigns for specific targets. The system performed successfully, obtaining profiles which fit in more than 60% of cases with the real user navigation sequences and in more than 90% of cases with the user interests. Moreover the automatically extracted semantic structure of the website and the interest profiles were validated by the BTw DMO staff, who found the knowledge provided to be very useful for the future.

Web 2.0 technologies have made it easily for Web users to create and spread objectionable text content, which has been shown harmful to Web users, especially young children. Although detection methods based on key word list are superior in achieving faster detection and lower memory consumption, they fail to detect text content that is objectionable in semantic description. A framework that can perfectly integrate semantic model and detection method is proposed to perform probability inference for detecting this kind of Web text content. Based on the observation that an objectionable scene could be described by a set of sentences, a topic model which is learnt from the set is employed to act as a semantic model of the objectionable scene. For a given sentence, probability value which shows the likelihood of the sentence with respect to the model is calculated in the framework. Then we use a mapping function to transform the probability value into a new indicator which is convenient for making final decision. Extensive comparison experiments on two real world text sets show that the framework can effectively recognize semantic objectionable text, and both the detection rate and the false alarm rate are superior to those of traditional methods.

Highlights? Application of latent topic models for search and recommendation in Enterprise Social Software. ? Generation and refinement of enterprise knowledge structures with latent topic models. ? Item to item collaborative and content based recommendations with Latent Dirichlet Allocation. Enterprise Social Software refers to open and flexible organizational systems and tools which utilize Web 2.0 technologies to stimulate participation through informal interactions. A challenge in Enterprise Social Software is to discover and maintain over time the knowledge structure of topics found relevant to the organization. Knowledge structures, ranging in formality from ontologies to folksonomies, support user activity by enabling users to categorize and retrieve information resources. In this paper we enhance the search and recommendation functionalities of Enterprise Social Software by extending their knowledge structures with the addition of underlying hidden topics which we discover using probabilistic topic models. We employ Latent Dirichlet Allocation in order to elicit hidden topics and use the latter to assess similarities in resource and tag recommendation as well as for the expansion of query results. As an application of our approach we have extended the search and recommendation facilities of an open source Enterprise Social Software system which we have deployed and evaluated in five knowledge-intensive small and medium enterprises.