Research On WeChat Cross-media Retrieval Based On Semantic Association

WeChat,as an instant messaging software that can achieve information acquisition,communication,retrieval and dissemination,is favored by the majority of mobile phone users.With the advent of the era of big data,WeChat data also gradually presented the characteristics of variety,huge amounts of sex and complexity,which contains rich multimedia data resources,such as text,images,audio,and video and so on,these heterogeneous multimedia data are widespread underlying characteristics,the characteristics of high-level semantics is rich,has brought the huge challenge to the traditional multimedia retrieval,flexible and can realize different types of data across the cross-media retrieval emerges as The Times require.How to mine and correlate the semantic features of multimedia data is the key problem of cross-media retrieval.At first,this paper using literature metrological methods published from 2004 to2018 cross-media retrieval related literature were analyzed,and the discussion of cross-media retrieval research background,research significance and research status quo on the basis of,according to documents released time,subject distribution,journals,institutions,the author,and keywords,excavation and research hotspot of cross-media retrieval development,in order to further promote the development of domestic related cross-media retrieval research provides reference significance.Aiming at the heterogeneity and semantic gap of WeChat text and image in the underlying features,this paper takes semantic association analysis as the breakthrough point and designs a cross-media retrieval model of WeChat based on semantic association.The model first to WeChat submitted by the user query examples feature analysis and extraction,text and image is studied on the content of the underlying characteristics of statistical relationship,at the same time in the feature dimension reduction maximum keep the potential relevance,use relevant technology mining text and image semantic characteristics,through the typical correlation analysis feature mapping,and stored in the cross-media semantic correlation subspace;Then,the Euclidean distance function is used to calculate the semantic correlation between textand image and realize cross-media retrieval between text and image.Finally,the cross-media semantic association subspace is updated in real time through relevant feedback strategy and user interaction behavior,so as to provide users with results more consistent with the retrieval requirements.The WeChat cross-media retrieval model based on semantic association can realize the mutual retrieval between text and image,and through the comparative analysis with cross-media retrieval results using the canonical correlation analysis,it proves that the method proposed in this paper is feasible and has good retrieval performance.