Kaisong Song

A word-emoticon mutual reinforcement ranking model for building sentiment lexicon from massive collection of microblogs

Recently, more and more researchers have focused on the problem of analyzing people’s sentiments and opinions in social media. The sentiment lexicon plays a crucial role in most sentiment analysis applications. However, the existing thesaurus based lexicon building methods suffer from the coverage problems when faced with the new words and new meanings in social media. On the other hand, the previous learning based methods usually need intensive expert efforts for annotating training datasets or designing extraction patterns. In this paper, we observe that the graphical emoticons are good natural sentiment labels for the corresponding microblog posts and a word-emoticon mutual reinforcement ranking model is proposed to learn the sentiment lexicon from the massive collection of microblog data. We integrate the emoticons and candidate sentiment words in the microblogs to construct a two-layer graph, on which a random walk is run for extracting the top ranked words as a sentiment lexicon. Extensive experiments were conducted on a benchmark dataset with various topics. The results validate the effectiveness of the proposed methods in building sentiment lexicon from microblog data.

Build Emotion Lexicon from Microblogs by Combining Effects of Seed Words and Emoticons in a Heterogeneous Graph

As an indispensable resource for emotion analysis, emotion lexicons have attracted increasing attention in recent years. Most existing methods focus on capturing the single emotional effect of words rather than the emotion distributions which are helpful to model multiple complex emotions in a subjective text. Meanwhile, automatic lexicon building methods are overly dependent on seed words but neglect the effect of emoticons which are natural graphical labels of fine-grained emotion. In this paper, we propose a novel emotion lexicon building framework that leverages both seed words and emoticons simultaneously to capture emotion distributions of candidate words more accurately. Our method overcomes the weakness of existing methods by combining the effects of both seed words and emoticons in a unified three-layer heterogeneous graph, in which a multi-label random walk (MLRW) algorithm is performed to strengthen the emotion distribution estimation. Experimental results on real-world data reveal that our constructed emotion lexicon achieves promising results for emotion classification compared to the state-of-the-art lexicons.

A Novel Approach Based on Multi-View Content Analysis and Semi-Supervised Enrichment for Movie Recommendation

Although many existing movie recommender systems have investigated recommendation based on information such as clicks and tags, much less efforts have been made to explore the multimedia content of movies, which has potential information for the elicitation of the user’s visual and musical preferences. In this paper, we explore the content from three media types (image, text, audio) and propose a novel multi-view semi-supervised movie recommendation method, which represents each media type as a view space for movies. The three views of movies are integrated to predict the rating values under the multi-view framework. Furthermore, our method considers the casual users who rate limited movies. The algorithm enriches the user profile with a semi-supervised way when there are only few rating histories. Experiments indicate that the multimedia content analysis reveals the user’s profile in a more comprehensive way. Different media types can be a complement to each other for movie recommendation. And the experimental results validate that our semi-supervised method can effectively enrich the user profile for recommendation with limited rating history.

Detecting Positive Opinion Leader Group from Forum

Forum has long been the main way of communication, and more and more users publish their opinions by it. The most influential users or opinion leaders will contribute to the formation of information, especially the positive influential users who can guide public opinions and make positive influence. Positive Opinion Leader Group (POLG) represents a group of users, each of who expresses the similar content and same sentiment orientation with their followers to a great extent, who are regarded as the most influential men during the information dissemination process. However, most existing researches pay less attention to the implicit relationship, heterogeneous structure and positive influence. In this paper, we focus on modeling multi-themes user network of forum with explicit and implicit links for this purpose. In detail, we put forward a data structure Longest Sequence Phrase Tree (LSP-Tree) for representing comments on forum, measuring the similarity between comments based on LSP-Tree to obtain implicit links, and further detecting positive opinion leader group. Experiments using dataset from Tianya forum show that our method can detect positive opinion leaders group effectively and efficiently.

Detecting Multiple Coexisting Emotions in Microblogs with Convolutional Neural Networks

Analyzing human sentiments and emotions is a critical problem in cognitive computing. One fundamental task of sentiment analysis is to infer the sentiment polarity or emotion category of subjective text, such as microblogs. Most existing methods treat sentiment classification as a type of single-label supervised learning problem that classifies a microblog according to sentiment polarity or a single-labeled emotion. However, multiple fine-grained emotions may coexist in a single tweet or sentence of a microblog. We regard emotion detection in microblogs as a multi-label classification problem. First, we develop a graph-based algorithm to automatically build emotion lexicons, which are further utilized to construct distant-supervised corpora from massive microblog datasets. Then, a ranking-based multi-label convolutional neural network model (RM-CNN) that considers the order and relevance of labels is proposed to address emotion detection in microblogs. The RM-CNN model is pre-trained using the distant-supervised corpus and then fine-tuned using specific training data without the need for any manually designed features. Extensive experiments on two real-world datasets demonstrate substantial improvements of our proposed RM-CNN model over the state-of-the-art baseline methods in terms of multi-label classification metrics. We propose an effective RM-CNN model with a distant-supervised learning framework for detecting multiple coexisting emotions in the short text of microblogs.

Build Emotion Lexicon from the Mood of Crowd via Topic-Assisted Joint Non-negative Matrix Factorization

In the research of building emotion lexicons, we witness the exploitation of crowd-sourced affective annotation given by readers of online news articles. Such approach ignores the relationship between topics and emotion expressions which are often closely correlated. We build an emotion lexicon by developing a novel joint non-negative matrix factorization model which not only incorporates crowd-annotated emotion labels of articles but also generates the lexicon using the topic-specific matrices obtained from the factorization process. We evaluate our lexicon via emotion classification on both benchmark and built-in-house datasets. Results demonstrate the high-quality of our lexicon.

CTROF: A Collaborative Tweet Ranking Framework for Online Personalized Recommendation

Current social media services like Twitter and Sina Weibo have become an indispensable platform, and provide a large number of real-time messages. However, users are often overwhelmed with large amounts of information delivered via their followees, and may miss out on much enjoyable or useful content. An information overload problem has troubled many users, especially those with many followees and thousands of tweets arriving every day. In this case, real-time personalized recommendation plays an extreme important role in microblog, which needs analyzing users’ preference and recommending most relevant and newest content. Both of them pose serious challenges. In this paper, we focus on personal online tweet recommendation and propose a Collaborative Tweet Ranking Online Framework (CTROF) for the recommendation, which has integrated the Optimized Collaborative Tweet Ranking model CTR+ and Reservoir Sampling algorithm together. The experiment conducted on a real dataset from Sina microblog shows good performance and our algorithm outperforms the other baseline methods.

Recommendation vs Sentiment Analysis: A Text-Driven Latent Factor Model for Rating Prediction with Cold-Start Awareness

Review rating prediction is an important research topic. The problem was approached from either the perspective of recommender systems (RS) or that of sentiment analysis (SA). Recent SA research using deep neural networks (DNNs) has realized the importance of user and product interaction for better interpreting the sentiment of reviews. However, the complexity of DNN models in terms of the scale of parameters is very high, and the performance is not always satisfying especially when user-product interaction is sparse. In this paper, we propose a simple, extensible RS-based model, called Text-driven Latent Factor Model (TLFM), to capture the semantics of reviews, user preferences and product characteristics by jointly optimizing two components, a user-specific LFM and a product-specific LFM, each of which decomposes text into a specific low-dimension representation. Furthermore, we address the cold-start issue by developing a novel Pairwise Rating Comparison strategy (PRC), which utilizes the difference between ratings on common user/product as supplementary information to calibrate parameter estimation. Experiments conducted on IMDB and Yelp datasets validate the advantage of our approach over state-of-the-art baseline methods.

Online Friends Recommendation Based on Geographic Trajectories and Social Relations

With the rapid development of GPS-enabled mobile devices, people like to publish online data with geographic information. The traditional online friend recommendation methods usually focus on the shared interests, topics or social network links, but neglect the more and more important geographic information. In this paper, we focus on users’ geographic trajectories that consisting of a series of positions in time order. We reduce the length of each trajectory by clustering the points and normalize every trajectory according to its positions and time in the trajectory. The similarity between trajectories is computed based on the distance of each corresponding point pair in the respective trajectory and the trajectories’ trends. The potential online friends are recommended based on the trajectory similarity and social network structures. Extensive experiment results have validated the feasibility and effectiveness of our proposed approach.