A novel content-based recommendation approach based on LDA topic modeling for literature recommendation

In an application such as literature recommendation, we require a comprehensive recommender model that can generate relevant recommendations similar to the literature provided in the input query. In this paper, we have proposed a novel content-based recommender system based on Latent Dirichlet Allocation (LDA) and Jensen-Shannon distance, which can be used specifically for the task of literature recommendations. We have compared this model with the standard cosine-similarity based approach for its use to generate scientific publication recommendations, in which recommend suitable journals/conferences to publish a research work based on the abstract of the user's manuscript as an input. We evaluated the results of both the proposed model and standard cosine-similarity based approach over unseen documents and achieved a precision score of 62.58% while the standard cosine-similarity based approach achieved a precision of only around 48%.

Automated summary

A novel content-based recommender system utilizing LDA and Jensen-Shannon distance was proposed for literature recommendations. Results showed a precision score of 62.58% on unseen documents, outperforming the standard cosine-similarity approach which achieved only around 48% precision.