基于神经网络Word2Vec的个性化推荐比较研究

Personalized recommendation refers to recommending personalized content for users based on their characteristics, historical behaviors, interests, needs, etc. With the rapid development of artificial intelligence technology, personalized recommendation has derived concepts such as algorithmic recommendation and intelligent recommendation that express the same meaning. Personalized recommendations have greatly alleviated the information overload of users, and have received extensive attention from the practice and academic circles. This paper combines Zhihu, a social Q&A platform in China, and CNKI, an academic platform, to make a comparative study of personalized recommendations. Firstly, the answers to the relevant questions and the abstracts of related journal literature were collected based on the subject words on Zhihu and CNKI respectively, and the corpora representing the practice and academic circles were formed. Then, the data cleaning and model training were carried out for the two corpora, and two neural network Word2vec models were constructed. Then, the top 20 words that are most semantically similar to "personalized recommendation", "algorithm recommendation", "intelligent recommendation" and "user" were combined to compare the hot spots of the two platforms. Finally, the word embeddings of the two platforms are compared by dimensionality reduction technology and visualization methods. Following the data-driven research paradigm and innovatively using the neural network Word2vec method, this paper finds that there are obvious differences between the social Q&A platform and the academic platform in terms of the focus and word embedding of personalized recommendation, which provides a reference for identifying the gap between the practice and academic circles, and promotes the research on personalized recommendation from a new perspective.