Acquisition,Expression And Mining Of Spatial-temporal Information In Tourism Cross-media Big Data

There is lots of unstructured spatial-temporal information in the cross-media big data from Internet,which is significant for personalized service based on the perception of tourism environment and tourist condition.In order to better acquire and use the spatial-temporal information,the thesis studied the acquisition,expression and mining for tourism cross-media big data.The main works of the thesis are as follows.(1)The thesis proposed a spatial-temporal information acquisition algorithm based on geo-tagging(GT-STA).The proposed algorithm built a spatial-temporal crawling model to extract information of place name,time and location,filtered noise data and acquired spatial-temporal information from geo-tagging.The thesis proposed a scenic spots' multi geographic features acquisition algorithm based on geo-tagged photo(GTP-MFA).The proposed algorithm used Mean Shift algorithm to divide a complex scenic spot to several inner spots,introduced user feature to calculate the weight of each geographic feature and realized the multi geographic features acquisition of complex spots.Experimental results show that the proposed GT-STA algorithm gains a promotion of 54.9%and 26.7%in term of precision compared with Named Entity Recognition based spatial-temporal information acquisition algorithm(NER-STA)and toponym database based spatial-temporal information acquisition algorithm(TD-STA),respectively.And the proposed GTP-MFA algorithm gains an average promotion of 20.0%and 23.3%in term of precision,and an average promotion of 59.8%and 9.9%in term of recall,compared with SM-MFA and MS-MFA algorithms,respectively.(2)The thesis proposed a spatial-temporal information expression algorithm based on cross-media summarization(CMS-STE).The proposed algorithm fused image similarity,tag information and image-tag similarity into a cross-media similarity matrix,improved the summarization quality by a scoring mechanism,clustered candidate images and tags using AP algorithm and chose image and tag summarization from the clusters,which realized the spatial-temporal information expression in the form of cross-media summarization.Experimental results show that the proposed CMS-STE algorithm gains an average promotion of 22.7%and 14.7%in term of image precision,16.9%and 9.1%in term of tag precision,and an average promotion of 40.1%and 33.8%in term of image cross-media correlation rate,14.8%and 11.0%in term of tag cross-media correlation rate,compared with summarization algorithms based on K-means and AP,respectively.(3)The thesis proposed a MapReduce based travel pattern mining algorithm(MR-TPM).The proposed algorithm clustered the geo-tagging to obtain tourist POIs,mined frequent travel patterns,reduced the load of nodes by partial frequent itemsets pruning and obtained the frequent itemsets of tourist routes between POIs.Experimental results show that the proposed MR-TPM algorithm gains an average reduction of 57.1%and 35.9%in term of running time compared with k times MapReduce based Apriori algorithm(MRKA)and 2 times MapReduce based Apriori algorithm(MRA),respectively.The proposed MR-TPM algorithm was used to mine tourists' frequent activity patterns in Beijing.(4)The thesis designed and implemented a system for the acquisition,expression and mining of spatial-temporal information in tourism cross-media big data.The system verifies the effectiveness of the proposed GT-STA,GTP-MFA,CMS-STE and MR-TPM algorithms.Verification results show that these proposed algorithms can well satisfy the need of the acquisition,expression and mining of spatial-temporal information in tourism cross-media big data and the system is fault tolerant and meets tourists' demand.