Tag Semantics Recommendation Algorithm Research Based On Distributed Representations Of Words

Encouraging users to generate content is the main tone on the context of Web 2.0. Tags, which are free-format keywords or terms created by users to describe content, are best able to categorize or index the information and to cope with the large amount of data in a timely manner. Users can also share tags to their friends. Despite the above advantages, social tags are free text and are not constrained by a controlled vocabulary and annotation guidelines, tags tend to be noisy and sparse. In order to improve these bad situations, tag recommenders emerged as the times require and has drawn so much attention from researchers and scientists that it has become a hot research field of recommendation systems.This paper focuses on the tag recommenders in the social tagging system. The work shows as follows:Firstly, this paper introduces the models of recommendation systems and the basic algorithms. And then summarizes the existing techniques about tags recommendation based on the perspective of their merits and demerits. We find a few research focus on the semantic tag similarities or their related work just can apply to the English language systems. So we propose an algorithm based on the distributed representations of words that both have semantic and collaborative advantages.Secondly, we use the Skip-gram algorithm to compute tag similarity. Compared with the current mainstream of Wu & Palmer’s concept similarity algorithm, our algorithm has two advantages:a) It has more extensively applicable scope, Wu & Palmer’s algorithms are limited to English; b) It has no problem of words out of dictionary. Wu & Palmer’s algorithm is based on WordNet semantic hierarchy tree, however the WordNet just include 155,287 words, cannot deal with the words of dictionary; c) It can represent idiomatic phrases. For example, phrases like "less than 300 rating" occurs frequently in tag system. However WordNet cannot figure it out because the meanings of several words cannot be easily combined to obtain the whole meaning of phrase.Thirdly, we propose a semantic tags recommendation algorithm based on distributed representations of words. It combined with neural network language model and Hungary algorithm so that can better solve the problem about similarity between tags and similarity between users. In the end, we validate the validity of our proposed algorithms through experiments and compared with present recommendation techniques, we observe good improvements in accuracy.