Search

To solve the problem of lacking user ratings and inaccuracy of user topics model for danmaku video website, this paper proposed a danmaku video recommendation algorithm combing topic model and collaborative filtering ( DRCIT). Firstly, it constructed an implicit rating matrix with user behavior, and obtained the similarity of project rating. Secondly, it established an LDA topic model to generate project topics and calculate the similarity of project topics. Thirdly, it generated a comprehensive project similarity, and calculated the user-project prediction score. Then, through sigmoid penalty, it established the new similarity matrix of project topics and got user-project topic similarity combined with user history behavior. Finally, it took user-project topic similarity as weight to multiply with prediction score to get the final prediction score. The comparison experiments show that the algorithm can get reasonable recommendation results and improve the accuracy of recommendation. [ABSTRACT FROM AUTHOR]

LDA is based on the bag-of-words, which simplifies the complexity of modeling, but makes the semantic coherence of topics poor, and text representation ability is not strong. To solve this problem, this paper proposed the semantic distribution similarity based topic model. This model used GPU (generalized Pólya urn) model to add word-word and document-topic semantic distribution similarity to guide topic modeling under the framework of EM (expectation maximization) algorithm, which weakened the effect of bag-of-words hypothesis on topics from the semantic association level. Experiments on four public datasets show that the semantic distribution similarity based topic model is superior to the currently popular topic modeling algorithms in terms of topic semantic coherence and text classification accuracy, and the model improves the convergence speed and topic accuracy. [ABSTRACT FROM AUTHOR]

In order to solve the problem that the recommendation list :is biased towards popular projects and with poor diversity, this paper proposed the ARIFDP ( aggregation recommendation algorithm for embedding item fatigue and diversity preference) algorithm. First, it analyzed the user's diversity preferences by using the historical feedback data of the user to derive the user, got the degree of diversification. And then it constructed a project fatigue function negatively related to the number of evaluations; eventually aggregated the matrix decomposition and the project fatigue function, and the inclusion of various propensity degrees adjusted the weight of the fatigue function of the project, and increased the probability of the proposed project. Experimental results show that the ARIFDP algorithm can effectively improve the diversity of recommendation results on the premise of ensuring accuracy. [ABSTRACT FROM AUTHOR]

Aiming at the increasingly rich multi language information resources and multi-label data in news reports and scientific literatures, in order to mining the relevance between languages and the correlation between data, this paper proposed labeled bilingual topic model, which was applied on cross-lingual text classification and label recommendation. First of all, it could assume that the keywords in the scientific literature are relevant to the abstract in same article. And then it extracted the keywords and regarded it as labels, and aligned the labels with topics in topic model, instantiated the " latent" topic. Secondly, this paper trained the abstracts in article through the proposed topic model. Finally, it classified the new documents by cross-lingual text classifier, and also recommended the labels. The experiment result show that micro-F1 measure reaches 94. 81 % in cross-lingual text classification task, and the recommended labels also reflects the sematic relevance with documents. [ABSTRACT FROM AUTHOR]

This paper proposed a collaborative topic regression model based on the preference for frequent topic sets to address the item-cold-start problem in academic paper recommendation. The algorithm took into account the user' s preference for research hotspots when selected academic papers, and used frequent topic sets to represent research hotspots. So, it expressed user' s preference for research hotspots as the user's preference for frequent topic sets. Firstly, it obtained the papers-topic probability distribution matrix through LDA algorithm and filter out the topics with higher probability. Then, the algorithm mined the frequently-occurring topic sets and gets the relationships between papers and frequent topic sets. Finally, it used the user's preference for frequent topic sets for the prediction of unknown scores. Experiments on CiteULike datasets show that the algorithm improves the recall, accuracy and RMSE over the matrix factorization model and the collaborative topic regression model. [ABSTRACT FROM AUTHOR]

In recent years, the explosive growth of the number of App brings difficulties for user to find a suitable App. There are many limitations of the traditional recommendation approaches when they are applied to App, such as the cold start problem and different choice bias on different types of applications. This paper proposed a personalized recommendation algorithm TGRW, which was based on topic grouping and random walk. Because users have different biases and underling choice factors on the App of different categories, TGRW firstly divided App into categories based on their description information. Then it constructed a triple tuple graphic model consisting of user, App group and App. By applying random walk to calculate the preferential probability of user to each App, it obtained the recommendation list. Extensive experiments on the real data set shows TGRW gains significant improvements on recommendation performances than other methods. [ABSTRACT FROM AUTHOR]

Since the traditional methods of micro-blog sentiment analysis ignore the influence of images, the special symbol information and the context semantic information, which leads to the accuracy rate of the emotion analysis method is not high. Therefore , this paper presented a micro-blog sentiment method, which was illustrated and based on transferred variable. Firstly, it paper constructed an unsupervised sentiment analysis model USAMTV, which was based on transferred variable, to analysis the distribution of emotion in the text. That was to say, the emotion and topic were categorized by introducing the sentimental transferred variable of conjunction and thematic transferred variable of reposting symbols. Then it introduced the image factors as emotional weight to decide the emotional distribution of the text. Finally, it obtained the overall sentimental orientation of the micro-blog. Compared with the JST model and the ASUM model, this proposed method had higher accuracy in the real-world test dataset. The experimental results show' that the proposed method can predict the emotional tendency of microblog more accurately. [ABSTRACT FROM AUTHOR]

The state-of-the-art studies started paying attention to comprehensively analyze geographical information,comment information and social information, but there is still insufficient information mining. To this end, this paper proposed a joint probabilistic generative model of multi-information fusion. Firstly, the framework learned interest topics of users and POIs through textual information,by exploiting the aggregated hierarchical Dirichlet process model,which could automatically learn the number of topics,to replace latent Dirichlet allocation model. Secondly ,according to the kernel density estimation method , whose bandwidth depended on the check-in distribution , the framework conducted personalized modeling of geographic information. Thirdly,it also took consideration of sequential patterns,which was the impact of the visited location to the non-visited location. Then , it modeled social relevance comprehensively. At last,based on the joint probabilistic generative model,this paper proposed the TGSS-PGM model, exploiting multi-type contextual information and incorporating these factors effectively. Experimental results in real world social net-work show that the proposed model outperforms state-of-the-art recommendation algorithms in terms of precision and rating error. [ABSTRACT FROM AUTHOR]

In the interest of solving the review honesty problems in current recommendation algorithms based on review analysis, and the data sparsity problems in traditional collaborative filtering algorithm, this paper involved the modified user preferences and user trust into traditional collaborative filtering algorithm, and proposed a new user-based collaborative filtering algorithm based on user ratings, reviews and user trust by analyzing the topic distributions and feedback information included in reviews from users. In detail, with the consideration of users’ reviews, it investigated the distribution of item features on different topics, and the level of preference of different users, to generate topic distribution of users’ reviews. To continue, this paper calculated the value of review diversity to magnify distinctive features in topic distribution, and generated the value of review helpful-degree by utilizing review feedback data to rectify user preferences and reduce the influences from spam reviews. On this basis, it further introduced user trust to enhance the accuracy of user similarity. The results of experiment on real dataset show that the proposed algorithm can improve the quality of prediction performance. [ABSTRACT FROM AUTHOR]

Latent Dirichlet allocation(LDA) topic model is an effective tool to process unstructured documents. But it is built on bag-of-words(BOW) model assumption, which regard each document as a combination of the word, neither the order relationship between documentsnor the order relationship between words is concerned. To improve current model's accuracy, this paper came up with the centroid-word based context topic model, this model was based on the theory that the topic of a word in a document had strong relationship of the word which near by. When calculating the topic distribution for each word, it regared the word as the center, extend before and after several words as the window, and then performed a calculation on each window. This approach would generate the corresponding order of each window, the same as the order of words, and because of the contexts of each word were different, so the distribution of each word had relationship with the location the word in the corresponding document. Experiments show that the centroid-word based context topic model has the better accuracy and convergence rate on unknown datasets. [ABSTRACT FROM AUTHOR]

In modern society, the micro-blog is an important social network. Repost is one of the most important features of micro-blog. However, the previous studies should have paid close attention to the motivation of repost and the interest-repost. Aiming at the feature of repost, the paper proposed an AT-IR model which was based on the original AT model tapping the interest-reposting to find the interest of micro-blog users. Besides , by means of crawling the micro-blog experimental data sets , this model is better than the AT model in the aspect of extraction accuracy. In addition , this method has shown the expansibility in the aspect of other subject model. [ABSTRACT FROM AUTHOR]

To solve the problem of content sparsity and lack of context information existed inherently in short texts, this paper proposed a biterm topic model (BTM) incorporating word vector features LF-BTM based on BTM. This model introuded latent feature model which utilized its abunnant word vector information to offset the data sparsity. Generation of words in each biterm was influenced jointly by topic-word multinomial distribution and latent features model in the improved generative process. Parameters in the model could be learned by of Gibbs sampling method. Experimental results on real-world short texts datasets demonstrate that the model can integrate word vectors trained from external general large-scale corpora to produce significant improvements on topic coherence. [ABSTRACT FROM AUTHOR]

Topic models based on bag-of-words got topics represented by individual words which were difficult to understand. This paper thus proposed using events as documents’ terms, meanwhile, considered Biterm topic model combined with GPU model which incorporated event correlation knowledge for inference, alleviated the adverse effect of event sparsity from co-occurrence and semantic two aspects. Experimental results show that the proposed method guarantees both interpretable and coherent topics. [ABSTRACT FROM AUTHOR]

In order to overcome the sparsity problem of traditional collaborative filtering algorithm, combining the characteristics of user ratings with user reviews, litis paper proposed user-based collaborative filtering with reviews and ratings, and item-based collaborative filtering with reviews and ratings. The algorithms utilized topic model to generate review topics distribution, and utilized ratings to generate review altitude. Then it used the review altitude to magnify the prominent aspects of review topics distribution, and finally predicted ratings and made recommendations. The experimental results show that the proposed algorithms can improve the quality of recommendation in sparse datasets. [ABSTRACT FROM AUTHOR]

The existed topic models are influenced by the characteristics of the short text in the social network when mine topics, which leads to the low quality of the topics and the small different between topics. To solve the problem, this paper proposed a new topic model U_BTM( user -bitem topic model) based on the topic model BTM. Firstly, it used K-means clustering algorithm to cluster the short text into a document, then modeled the generation schema of bitem according to the users’ topic, and adopted the Gibbs sampling method to derive the model, finally obtained the potential topics and the topic distribution of users. Experimental results show that the U_BTM model can not only get the potential topics and the topic distribution of each user, hut also has a higher quality and difference and lower perplexity than other models. [ABSTRACT FROM AUTHOR]

Aiming al the problem of the information extraction of history evolution, this paper proposed a method combined PAM with topical TextRank . PAM could fetch the distribution of different topics and the distribution of topics and words. Based on the distribution modeled by the PAM. topical TextRank could extract the keywords which were more relevant to historical evolution topic. Because historical evolution topic was complex and highly relevant to other topics. W hether the words were relevant to the historical evolution topic was not clear and hardly confirmed. So. PAM got both the distribution of topics and the distribution of topics and words, which was very useful to solve those problems. Furthermore, the historical evolution topic facilitated topical TextRank to extract the words which were more relevant to historical evolution topic. The result shows that the method combined the PAM and topical TextRank is more effective to extract the topic information of history evolution. [ABSTRACT FROM AUTHOR]

Aiming al the retrieval semantic information in Mongolian, this paper proposed a new method combined topic model latent dirichlet allocation (LDA) and language model. This method modeled Mongolian documents with LDA and language model, estimated parameters with Gibbs sampling and represented probability of word, it could mine the hidden relationship between the different topics and the words from documents, got the topic distribution and computed the similarity of keywords topics. Finally, it returned to the most relevant documents with topics. Experimental results show that the method has a higher performance in topic semantic compared with one sole model. [ABSTRACT FROM AUTHOR]

In text streaming data, documents of text stream into system continuously, and huge amount and quick update velocity of data pose great challenges for discovering and tracking topics in such text streaming. In order to handle those prob-lems, this paper proposed a collective matrix factorization based topic discovering and tracking model. In order to deal with mas-sive data, it partitioned the streaming data into batches, and in order to track the evolution of topics, it assumed the current topic distribution as the linear conversion of the previous one. This paper applied collective matrix factorization to represent the cur-rent data as both current topic distribution and previous topic distribution, and got the current topic distribution via factoring both matrices. During the process of solving the optimization problem, it got the parameter update strategy by the Karush-Kuhn-Tucker conditions, and gave the corresponding solving algorithm. During the experimental comparison with Yahoo dataset, the proposed algorithm can better discover topic distribution in text streaming than related works, and also can track the evolution of topics as time going. [ABSTRACT FROM AUTHOR]

Aiming at the problems of uneven students' abilities and the absence of specificity in MOOC teaching process, this paper presented a novel topic interaction model based on the local community detection. It modeled the evaluation process of students' ability as hierarchical structure of a topic model. It introduced the local community detection algorithm to reasonably classify various abilities. Meanwhile, based on the original course experiment platform proposed before, it established an improved MOOC platform to widely trace the behavior information and provided with highly interaction. Experimental results on the teaching practice of digital signal processing course show that topic interaction model achieves higher accuracy rates compared with traditional assessment methods. Supplemented with specific communication and guidance, students' abilities can be enhanced effectively to some degree. [ABSTRACT FROM AUTHOR]

Traditional topic model assumes that each document is related only one topic, but in practice, a document is usually related with multi topics. This paper represented a document with combination of multi topics with the LDA model, and proposed a LDA based mixture model for Ad hoc information retrieval method in language model framework. The proposed method combined the LDA and the document model. Compared with cluster-based model, it has a higher performance, while keeping low computation complexity, and thus can be used in information retrieval of large scale document collections. [ABSTRACT FROM AUTHOR]

Because the BTM mining the topic is very slowly, so the BTM can not do very well on the large scale mircoblog. S o this paper presented a distributed BTM that updated the gobal topic word matrix by sampling the local host biterm set which caHed PBTM. It used multifile hosts Gibbs sample on the local biterm set and then updated the gobal topic word matrix after each Gibbs sample ' s iteration until sampling convergence. And this paper implemented PBTM algorithm through the MPI cluster. Experimental results show that PBTM topic mining' s speed is great faster than the BTM. [ABSTRACT FROM AUTHOR]

General traffic classification methodologies cannot describe the detail information of network communication. This paper proposed a novel method based on users' online activities to classify traffic. This method described different traffic with characteristics of activity status, and combined vector quantization with topic model to extract characteristics of activity status from traffic. Then it built classification model and identified traffic with these characteristics by some machine learning algorithms. The experimental results show that category of users' activities give a more elaborate scheme to traffic classification. And with the same machine laming algorithm, the recognition rate of classification based on characteristics of activity status is better than traditional ways' rate. [ABSTRACT FROM AUTHOR]

As an important research direction in text mining, this paper widely studied unsupervised and weak supervised generative models in text sentiment classification. It described the definition, classification and development of the generative model in text sentiment classification, focused on the relationship between the different generative sentiment models, with introduced the most representative models. Furthermore, it summarized the trends of the generative model and development in the future. [ABSTRACT FROM AUTHOR]

In topic models of the social networks, these are usually some frequent words, such as Barack Obama, and they appear in many different topics. Users usually are interested in these words, so they cannot be removed while analyzing, and this poses a huge challenge for topic analysis. In order to solve this problem, this paper took popularity into consideration while modeling topics. This paper defined the concept of popularity for nodes in the social networks, and proposed an extended topic model containing node popularity. The experiments show that, the proposed model has better prediction ability, and is more effective in prediction accuracy than related works. [ABSTRACT FROM AUTHOR]