Semantic Recognition Of Individual Activities Based On Social Media Data

The popularity of the Internet and multi-functional mobile devices provides more solutions for the data extraction of mobile objects,the data types and data volumes are more abundant.Social media data records the activities and moving trajectories of individuals in the real world from multiple perspectives,and with the improvement of sensor accuracy,the data becomes more authentic and authoritative.Location-based social networking service platforms include Weibo,We Chat,Four Square,Twitter,etc.Many users share news on these platforms every day,including time,location,and other information in the form of pictures,videos,and text.These information directly or indirectly reflect individual activities,which are closely related to urban transportation,public services,and infrastructure,etc.The identification of individual activity behaviors is of great value for the construction of user profiles,accurate recommendations,customized services,and group mobility analysis.It can also support the decision-making of urban planning and construction.This paper proposes a method that identifies individual activity semantics with social media check-in data.The time,location,and text information of the check-in data describe the individual's behavior and activities.Existing research on semantic recognition of individual activities mainly focuses on the exploration of temporal and spatial features,and lacks the mining of textual semantic information.In order to solve these problems,this paper proposes a multi-feature fusion semantic recognition algorithm for individual activities.In addition to the temporal and spatial characteristics,this paper further excavates the spatial activity preference features and the textual semantic features of check-in place names from the perspective of group preference and individual preference..The main contents of this paper are summarized as below:1.Spatio-temporal feature extraction of social media check-in data.The method respectively extract activity-related features from time and location information.The periodic and sequential information of activities and behaviors contained in the time information is described by the information of season,workdays or weekends,and the specific time.Spatial activity preference characteristics can be mined with spatial information.For hot spots in space areas,the spatial preference quantification algorithm is deployed to quantify the spatial preference of activity behavior,then represent the features perceived by vision or perception as numerical values.2.Textual semantic feature extraction of social media check-in data.The names of the check-in locations is closely related to the attributes of the place,which can reflect the behavior characteristics of individual activities.This paper excavates the semantic information of the text from the check-in location names,and extract the text semantics of the check-in location name with the bidirectional deep self-attention network model.The vector representation of the hidden activity semantic information in the text is utilized to complete the semantic vector conversion of the word.3.Construction of individual activity semantic recognition model classifier.This paper compares the classification and recognition capabilities of extreme gradient boosting model,support vector machine,random forest model and K-nearest neighbor algorithm for multifeature fusion feature vectors.Finally,the best performing classifier is selected.The multi-feature fusion method of individual activity semantic recognition proposed in this paper achieves the accuracy of 87.6% in individual activity semantic recognition.Experiments have proved the effectiveness of the features extracted in this paper.Compared with related algorithms,the algorithm in this paper also has better performance.