An adaptive hybrid deep learning architecture for sentiment analysis - based recommendations on social networks

Resumen

With the explosion of Web 2.0 and the rise of blogs, forums, and online social networks, different opinions about a particular topic can be easily found from millions of users on these websites. For example, users discuss current experiences, share their points of view on specific facts, and offer praise or complaints about specific products they have just bought. This kind of information plays a key role in various applications, such as to track comments or reviews of customers for recommender systems, and to analyze surveys that an organization itself conducts. The problem of automatically extracting opinions from online user-generated texts, known as “opinion mining” or “sentiment analysis,” has been a growing research topic recently. The study of public opinion can provide us with valuable information. The analysis of sentiment on social networks has become a powerful means of learning about the users’ opinions and has a wide range of applications. However, the efficiency and accuracy of sentiment analysis is being hindered by the challenges encountered in natural language processing. In recent years, it has been demonstrated that deep learning models are a promising solution to these challenges. In addition, sentiment analysis can be beneficial to recommender systems. As well, a growing body of research examines how sentiment analysis is being applied in recommender systems. Such an analysis can improve the understanding of users’ attitudes, opinions, and emotions, which is beneficial to integrate it into recommender systems for achieving higher recommendation reliability. Social media data has been exploited in different ways to address some problems, especially those associated with Collaborative Filtering (CF) methods. Sparsity and gray-sheep problems are two of the main reasons CF methods do not provide the reliability required in some recommender systems. In particular, when only sparse ratings data is available to a recommender system, sentiment analysis can play a key role in improving recommendation quality. This is due to the fact that recommendation algorithms mostly rely on users’ ratings on items to select the items to recommend. Such ratings are usually insufficient and very limited. On the other hand, sentiment ratings of items that can be derived from online news services, blogs, social media, or even from the recommender systems themselves are seen as capable of providing better recommendations to users. Sentiment-based models have been exploited in recommender systems to overcome the data-sparsity problem that exists in conventional recommender systems. This thesis addresses that gap by means of a comprehensive comparison of sentiment analysis methods in the literature and by an experimental study to evaluate the performance of deep learning models and related techniques on datasets about different topics. The research question aims to determine whether it is possible to present outperforming methods for multiple types and sizes of datasets. Another question raised in this thesis is whether hybrid models perform better than single models regardless of the characteristics of the datasets. Therefore, the aim of our work is the proposal of hybrid models and the study of their behavior with different types of datasets from different domains. Then, we present an approach to use sentiment analysis in recommender systems, in wich user opinions and explicits ratings are combined to provide recomendations. This application is based on an adaptive recommender system architecture, some techniques for feature extraction, and deep learning models for sentiment analysis. Hence, integrating sentiment in recommender systems may significantly enhance the recommendation quality of these systems. We applied deep learning models with TF-IDF and word embedding to eight datasets, including tweets and reviews. We implemented the state-of-the-art sentiment analysis approaches based on deep learning, and combined models to increase the accuracy of sentiment analysis. In addition, we evaluated methods for integrating sentiment analysis into general recommender systems for streaming services on popular public review datasets. The experimental results show that the proposed approach significantly improves the recommender system performance.