Research On User Interest Model And Real-time Algorithm Of Personalization Recommendation

With the development of web technology, tagging system becomes well-known by people as anew form of interaction. A user in the system can create a new tag or use an existing tag to mark anyitem. In recent years, researchers have found that tags can describe both users’ interests and thecharacteristics of items. However, due to the difference of personal understanding, there are stillsome drawbacks such as semantic fuzzy, ambiguous, and tag abuse. The thesis proposes a set ofmethods to standardize tags, map user’s tags to standard tags; Experiments show that the semanticproblem is solved to a certain extent.The interests of the users are divided into long-term interest and short-term interest, long-terminterests can abstract by tags and short-term interest can get by browsing behavior of the user.Standard tags are used to establish the stable interest model and immediate interest the model, andboth of models are combined to establish a real-time user interest model. Meanwhile, the thesisproposes a model update method based on natural forgotten in order to express the user’s currentinterests and its degree accurately.Experiments indicate the update method improve the accuracy ofmodel effectively.Personalized recommendation algorithm as a core of a recommendation system is closelyrelated to the recommendation quality of the recommendation system. The thesis proposes animproved Slope one recommendation algorithm, M-Slope one. The algorithm calculates thesimilarity between users based on the real-time user interest model, establishes a nearest neighborgroup for each user by the similarity to reduce the search range for related items, and meanwhileimproves the rating average deviation formula of items with the similarity. Experiments show thealgorithm improves the accuracy of the algorithm recommended effectively.