Recommender systems using sentiment analysis
View/
Abstract
In recent years, a variety of recommender systems have been developed in order to meet business needs. Businesses aim in using valuable information such as user preferences and item similarities to recommend clients more and more relevant products. This process is well-known by now and it is used in a range of occasions.
Sentiment analysis (known as opinion mining) is a field within Natural Language Processing that builds systems that try to identify and extract opinions within text. Usually, besides identifying the opinion, these systems extract attributes of the expression e.g.: Polarity (if the speaker expresses a positive/negative opinion), Subject (the subject of a conversation) and Opinion holder (the person or entity that expresses the opinion). Currently, sentiment analysis is used in order to automate some human procedures like the above. Another use is taking advantage of this automation, and try to educe the expressing opinions that exist in many sites, forums and social media. With the help of sentiment analysis, this unstructured information could be automatically transformed into useful information for commercial application, product reviews, product feedback, and customer service.
For the purposes of this thesis, a recommendation system enhanced with sentiment analysis is built. Firstly, a sentiment analysis model was designed in order to be able to assign to a movie review a star rated from one to five. Secondly, an item-item collaborative filtering recommendation system was developed. Then the two of them were combined in order to study if the sentiment analysis could enhance or even replace the rating of the user in recommender systems. This thesis also investigates many types of recommendation systems and states the pros and cons of each one. Also, the sentiment analysis model as well as the recommendation system are measured and the measurements are analyzed.
The objective of this thesis is to investigate the degree of the enhancement that sentiment analysis offers to recommendation systems. This implementation of recommender systems boosted with sentiment analysis tries to explore the meaning and contribution with measurements and examples of opinion mining to recommendation systems. It also investigates whether extracting opinions from users steers the same users to get valuable information about future recommendations.