Research On Friend Recommendation Based On Fusion Of Users' Text Semantic And Sentiment Analysis

With the emergence of Web 2.0 technology and development of Internet, the SNS has been widely used in our real life. As the friend recommendation can enhance the stickiness of users, it becomes an important technology in the online social networks. Almost all SNSs have the function of friend recommendation, and in general use the users' static characteristics. However, the interests and emotions of users are often varied in their real life so that they can't describe the users' characteristics very well and the results are not good. On the contrary, the text of micro-blog contains many rich information of users'. In view of this, this thesis start the analysis and research on the users' characteristics from the semantic analysis and sentiment analysis of the text of micro-blog respectively in this thesis.Firstly, the text semantic of micro-blog and degree adverbs are considered and then a two stages method of friend recommendation is proposed. In the model, the key point is to use the text semantic technology to change the key words to compute the similarity of friends. Due to the characteristics of the users are changed by time, a time factor is also introduced into the computation of users' similarity. Then the thesis take further consideration on the user's emotional characteristics to compute the users' similarity through analyzing the emotional words in micro-blog text and get the final results. The fusion method of the text semantic analysis and degree adverbs analysis are better than traditional methods.Secondly, further deeper analysis is considered on the text and sentiment analysis based on the above research and then a friend recommendation method based on fusion of users' text semantic and sentiment analysis is proposed. Considering the different time that the user accesses to the information, the cross similarity calculation method are used to calculate the text similarity between users. In the research of text sentiment analysis, the degree adverbs can show emotional intensity and the negative words can change the tendency of emotion. So the users' emotional tendency with the effect of degree adverbs and negative words with the sentiment analysis of micro-blog of users are considered. The method is compared with some traditional methods on real datasets and the results show that the recommendation effectiveness is increased on various indicators.Finally, due to the influence of time decay, the recommendation results of SNSs are different from different types of users in different periods. The text of micro-blog of different period in this thesis are analyzed and then a friend recommendation system merged text semantic with sentiment analysis and time factor are designed.