Research On Affective Analysis And Recommendation Based On Deep Learning

With the increasing popularity of online consumption,people are gradually accustomed to browsing and screening products on the e-commerce platform.The evaluation information of the products contains the consumption experience and emotional attitude of the users in the past,and can provide more detailed decision-making basis for other users.The potential information mining of the evaluation data in a reasonable way can obtain the user's emotional attitude,and can also obtain relevant information such as the user's criticality and preference.However,as the number of Internet users increases year by year,the current amount of evaluation data is more and more huge,the content of evaluation is becoming more and more mixed,and the time cost of commodity screening based on evaluation information is increasing.One-by-one reading will not provide users with a more efficient basis for decision-making,so it is necessary to use automatic comprehensive affective analysis to identify and classify affective for massive evaluation data.In this paper,the deep learning algorithm is used to analyze the affective of the food stores comment text.Based on the affective analysis of the traditional RNN model,the attention mechanism is introduced and the structure of the model is improved.In order to solve the problem of many-to-one structure of RNN model in affective analysis and the problem of gradient disappearance and gradient explosion,a multi-index weighted RNN affective analysis(RNN-MWAA)model is proposed,which introduces the attention mechanism into the model.Based on the structure optimization of many-to-one output model,the emotional scores of each evaluation index are obtained.At last,each score is weighted according to different attention parameters to output the whole evaluation classification.A recommendation algorithm for synthesizing multi-index information is proposed.Combined with Word2Vec model and Euclidean distance formula,the evaluation classification of shops under different indexes is obtained by using the results of RNN-MWAA model analysis,and the users undefined preference information is extracted from the evaluation text to cluster similar users.The recommendation algorithm is constructed by using the above shop classification and user clustering information,and the feasibility recommendation content is added by the combination of geographic information and collaborative filtering algorithm.Design and development of emotional analysis and shop recommendation system.Based on the proposed RNN-MWAA model and integrated multi-source information recommendation algorithm,the system is designed in detail from the data layer,processing layer and application layer,the emotional analysis results and recommendation shops are visualized in the system.In this paper,the emotional analysis of Meituan evaluation data is carried out by using the RNN-MWAA model.It is shown that the analysis effect of the model is better than that of the MLP and LSTM models in terms of precision,recall rate and F1 index.The integrated multi-index information recommendation algorithm matches the index high-score shop with the user preference,and can realize personalized recommendation in the high-quality shop.Verified by the MAE index,the proposed recommendation algorithm has the reliability.Affective analysis and shop recommendation system realize the functions of Affective analysis interaction,shop recommendation,evaluation statistics and so on.The input evaluation text can be used for emotional analysis and output index score and attitude tendency.Using store attributes and user preference information to help users achieve efficient decision-making,at the same time,it provides a way for merchants to get user feedback quickly.