Fuzzy Text Search And Recommendation System Based On Word Embedding

Nowadays,as the Internet technology flourished,data information shows the trend of increasing exponentially.How to quickly and efficiently screen out the required information from the massive data has become an urgent problem to be solved.On the other hand,due to the continuous breakthrough of artificial intelligence technology,the search recommender system has been widely used in people's life as a convenient way to obtain information.However,there is a big gap between the existing search recommender system and the real sense of artificial intelligence,especially when the amount of data is huge while the computing power is limited,the recommend results quality and computing efficiency are often unsatisfactory.In the traditional search engine,using keyword matching to get search results,the computational efficiency is extremely poor,and usually lacks of semantic understanding,and when the search text input by the user were expressed fuzzily,it is difficult to capture the real needs of users,let alone obtain satisfactory results;the existing mainstream recommendation system also has a series of problems,including problems of cold-to-start,slow learning speed and huge matrix data need to be stored,etc.Besides,the user's personal characteristics,background,interest and other factors haven't been considered,which makes personalized recommendations for users becomes impossible.Addressing the above problems,this paper carries out the research works as follows:(1)Based on the research of existing search algorithms,a fast fuzzy search algorithm based on inverse filtering idea is proposed.The algorithm can quickly filter out the reasonable label by reverse filtering,so that the search efficiency was greatly enhanced.And the intelligent understanding of fuzzy text searched by users is realized effectively.The idea of fuzzy matching algorithm and the process of implementation are described in detail.Finally,the performance of the algorithm is verified by experiments.(2)Analyzed the existing mainstream recommendation algorithms,addressing the problems existing in the classic collaborative filtering recommendation algorithm,a user interest prediction recommendation system based on Bayesian network is proposed,which successfully avoids the problems existing in the collaborative filtering recommendation system including problems of cold-to-start,slow learning speed and huge data need to be stored,etc.The system uses the user's personal information,such as personal characteristics,personality,background,etc.,to predict users' really interests,which allows the system providing personalized resources recommendation services.Meanwhile,the adaptive feedback module is added to the system,and the network parameters can be adjusted according to the user's actual situation,thus the prediction accuracy of the recommendation system can be improved.(3)Studied and did some practice related to the deep learning word embedding tool word2vec,using the deep learning language model to train the word vector can achieve the semantic representation of the word,thus the system can understand the deep semantic information contained in the search text.Besides,based on the existing word2vec synonym association function,a keywords directional association algorithm was proposed as an improvement,this algorithm divides the high dimensional word vector space,thus the semantic mapping between user interest and resource keywords can be built,so that the potential source information users interested in can be achieved.Finally,the rationality and efficiency of the algorithm is proved by experiments.