Online Social Network Cross-media Search Based On User Search Intention Understanding

With the increasing popularity of online social networks,more and more users publish information,share life status,and search interesting topics through social network platforms.Online social networks have accumulated huge amounts of cross-media data,and these cross-media big data have strong dynamic,the context sparsity and cross-media semantic gaps,and the user's search intention is difficult to understand for social network.The above problems bring severe challenges and obstacles to cross-media search for online social networks.Therefore,it is of great significance to study online social network cross-media search based on user search intention understanding.In view of the above problems,this dissertation studies the theory and technology in the following four aspects:online social networks topic expression of cross-media information,the understanding and mining of user search intentions for online social networks based on user aggregation,the bursty topics discovery for online social networks based on sparse topic model and online social networks cross-media search based on user search intention understand.The main research results and innovations are as follows:(1)Aiming at the problem that current text topic expression algorithm cannot overcome the sparsity of context and cannot achieve dynamic topic expression,this dissertation proposes an online social network text topic expression algorithm(SCTE)based on dynamic self-aggregation topic model and constructs a dynamic self-aggregation topic model(SADTM).By aggregating short text into long documents,the sparsity problem of social network context is alleviated.The dynamic expression of text topic for social network is realized by deriving the current topic distribution in conjunction with previously learned topic distributions.Aiming at the problem that the current image representation algorithm ignores the central feature,an image topic expression algorithm(CAIE)based on the complementary attention mechanism is proposed.By constructing the complementary attention mechanism,the focused and consistent image features are obtained,which realizes the expression of online social network image topic.The experiment verifies the validity of the proposed SCTE algorithm and CAIE algorithm in the topic expression.(2)Aiming at the problem that current user search intention understanding methods require specific privacy data,which is not universal,this dissertation proposes an online social network user search intention understanding and mining algorithm(UAIU)based on user aggregation and constructs an online social network user aggregation topic model(UATM),and the distribution of users' search intentions is obtained.By using RNN and IDF to construct weight prior and distinguish modeling topic words from common words,this dissertation realizes the word relationship learning for online social networks.Combining user search intention distribution and follower intention distribution,the understanding and mining of online social network user search intention are realized without any privacy data,which makes the algorithm more universal.Experiments verify the effectiveness of the proposed online social network user search intention understanding and mining algorithm(UAIU)based on user aggregation.(3)Aiming at the problem that the current bursty topic discovery algorithm cannot automatically discover bursty topics in online social networks,this dissertation proposes an online social network bursty topic discovery algorithm based on sparse topic model(SBTD),combining the smoothing and weak smoothing prior in "Spike and Slab" prior,a sparse topic model(SRTM)based on "Spike and Slab" prior is constructed.The distribution of bursty topics and words in online social networks are obtained,and the focus of bursty topics is realized.The source of the topic is determined by the binary switch variable,and combining the obtained bursty topics distribution and words distribution,the automatic discovery of bursty topics for online social networks is realized.Based on Sina Weibo dataset,a number of comparative experiments are designed.The experimental results verify the effectiveness of the proposed SBTD algorithm.(4)Aiming at the semantic gap faced by online social network cross-media search process and the difficulty of understanding user's search intention,this dissertation proposes an online social network cross-media search algorithm(UCMS)based on user search intention understanding.Adopting the proposed online social network user search intention understanding and mining algorithm based on user aggregation in Chapter 3 to guide the learning of text semantics,a complementary attention mechanismbis established,which realizes the consistency association of cross-media data for online social networks.An online social network cross-media adversarial learning process is constructed,which obtains the semantically consistent representation.The experimental results demonstrate the effectiveness of the proposed online social network cross-media search algorithm(UCMS)based on user search intention.(5)By comprehensively utilizing SCTE,CAIE,UAIU,SBTD and UCMS algorithms proposed in this dissertation,a online social networks cross-media search system based on user search intention understanding is designed and implemented.The system includes cross-media information topic expression module,user search intention understanding and mining module,online social network bursty topic discovery module and online social network cross-media search module.