Research And Application Of Interdisciplinary Collaboration Recommendation Method For Scholars In University

With the development of science,it is not always easy to solve increasingly complex scientific problems with the knowledge of a single discipline,and interdisciplinary collaboration has become an important means to solve complex scientific problems and promote scientific innovation.As a research institution that brings together scholars from various disciplines,universities have an advantage in promoting interdisciplinary collaboration.Meanwhile,it is beneficial for the development of interdisciplinary disciplines to recommend suitable interdisciplinary collaborators to the scholars within the university.The interdisciplinary collaboration recommendation is one of the academic collaboration recommendation problems.The issue of the academic collaboration recommendation focuses on finding scholars who are highly similar.However,the interdisciplinary collaboration recommendation is different from the regular ones.Similarity among scholars is one of the most important targets considered,but the difference between disciplines will reduce it among potential interdisciplinary collaborators.It is challenging to keep the similarity between scholars while maintaining a certain degree of cross-disciplinary diversity when recommending interdisciplinary collaborators.The current method mostly recommends scholars in other fields that can collaborate with their current research fields.which comes from topic modeling or academic network structure similarity,and whether the research fields can collaborate with each other is mostly based on a priori knowledge.However,interdisciplinary collaboration is not always carried out in some known domains,and disregard of scholar similarity can lead to ineffective collaboration and communication due to the gap between scholars’ disciplinary backgrounds.Therefore,this thesis obtains a representation vector of information about scholars’ research domains based on network representation learning and pre-trained language models.The vector is used for overlapping community detection to recognize the research areas of scholars,then recommend interdisciplinary collaborators among scholars in several communities associated by the overlapping part of the community.The main work is as follows:(1)Analysis of scholars’ academic relationships within universities.Based on the data of university dissertation results,the academic cooperation network based on "scholar-paper" relationship and the co-journal network based on "scholar-journal" relationship were constructed,and the results of community detection on these two networks were analyzed to investigate whether the results of community detection on these two networks could reflect the association between scholars and research fields.We also analyzed the results of community discovery on the two networks to investigate whether the results of community discovery on the two networks reflect the association of scholars and research fields.We found that the results of community discovery in the collaborative network were more consistent with intra-college collaboration in the same field,while the co-journal relationship was a more objective representation of the relationship between scholars and their research fields.(2)Scholar representation vector acquisition.Firstly,we construct a heterogeneous information network of scholars-papers-journals,then use the metapath2 vec model to obtain the heterogeneous network structure embeddings,secondly,use the Sentence-Bert model to obtain the text embeddings of the papers of scholars,and then combine the two vectors to obtain the scholar representation vector,and analyze the similarity between the results of community detection based on different representation vectors and the real community.We find that the representation vectors combining network structure embeddings and text embeddings are more effective in identifying interdisciplinary collaborative groups.(3)Propose an interdisciplinary collaborator recommendation algorithm.First,a network of scholars is established and the vectors corresponding to the scholar nodes are transformed to the connected edges;second,the k-means algorithm is used to divide the connected communities,and the number of clusters is subject to the starting place of the multi-objective optimization algorithm,which is determined by the silhouette scores and the Davies-Bouldin Index;then the non-overlapping edges communities are transformed into overlapping communities of nodes;finally,the interdisciplinary collaboration is performed by then passing through the communities associated with the overlapping communities recommendation.The algorithm is able to recommend collaborators with appropriate disciplinary span for scholars without knowing which fields are available for collaboration.