Research On Personalized Recommendation Method For Short Text Emotional Analysis

With the rapid development of network technology,the explosive growth of data volume has brought problems to the screening of people's information.Therefore,some researchers have proposed to use the recommendation system to get more effectively extract valuable information.The purpose of the recommendation system is aimd to help people dig out their potential needs from a large number of information and eliminate the amount of redundant information.Original recommendation system has achieved certain results,but this recommendation is not targeted.These results are difficult to meet their wishes for people with high standard of demand.Therefore,with the improvement of user requirements standards,people need a more accurate personalized recommendation system than before.In the field of e-commerce,recommender systems are generally recommended by mining and analyzing the explicit behavior and implicit behavior of the user.Explicit behavior generally includes user's direct evaluation of goods marking,while implicit behavior mainly includes indirect information such as browsing information,collection records,reviews,contextual information and so on.Because comment information appears in the form of short text,some people analyze and comment short texts,extract characteristic words to build evaluation labels.It is convenient to users form intuitive perception of products.However,it is not effective to comment on the potential emotions in short text.Therefore,this paper proposes a recommended method oriented analysis of short text sentiment personalized.According to several common Chinese dictionaries,we use tire tree and use word segmentation model to mine users' comments on products in each fraction segment and the emotional words in short texts.Geting the emotional feature words in the text,counted the main related feature words and their weights in each score,and constructed the general feature emotional lexicon and satisfy the user's personalized emotional feature lexicon.Then we use the Naive Bayes method based on cosine similarity and combine the user personalized sentiment lexicon and general emotion lexicon to assign corresponding scores according to the distribution of feature words.Finally,built the user-Personalized emotional project scoring matrix,and provided the recommendation.At last,this paper use crawler technology to capture the reviews of clothing items on an e-commerce site in China as an experimental data set.A personalized recommendation method,a collaborative recommendation method and a project based recommendation method for short text emotional analysis are compared,the results showed that the constructed by mining comments essay emotion words the user personalized recommendation item score matrix can effectively enhance the effectiveness of the recommendation,to verify the effectiveness of the method.