A SERENDIPITOUS RESEARCH PAPER RECOMMENDER SYSTEM

In recent times, the rate at which research papers are being processed and shared all over the internet has tremendously increased leading to information overload. Tools such as academic search engines and recommender systems have lately been adopted to help the overwhelmed researchers make right decisions regarding using, downloading and managing these millions of available research paper articles. The aim of this research is to model a spontaneous research paper recommender system that recommends serendipitous research papers from two large and normally mismatched information spaces using Bisociative Information Networks (BisoNets). Set and graph theory methods were employed to model the problem, whereas text mining methodologies were used to process textual data which was used in developing nodes and links of the BisoNets graph. Nodes were constructed from weighty keywords while links between these nodes were established through weightings determined from the co-occurrence of corresponding keywords originating from both domains. Final results from our experiments ascertain the presence of latent relationships between the two habitually incompatible domains of magnesium and migraine. Word clouds indicated that there was no obvious relationship between the two domains, but statistical significance investigations on the terms indicated the presence of very strong associations that formed information networks. The strongest links in the established information networks were further exploited to show bisociations between the two habitually incompatible matrices. BisoNets were consequently constructed, exposing terms and concepts from two discordant domains that were bisociated. These terms and concepts were utilised in querying the one domain for recommendations in another domain. Hence, serendipitous recommendations were made since our bisociative knowledge discovery methodologies revealed hidden relationships between research papers from diverse domains. Finally, it was postulated that latent relationships exist between two incompatible domains, and when well exploited, it leads to the discovery of new information and knowledge that is useful to researchers in various fields, especially those engaged in multi-disciplinary research. Further research is being conducted to identify outlier linkers and connectors between domains of diverse subjects.