In order to improve the accuracy and usefulness of the churn prediction model, the core elements of the research content were designed to include collecting data on customer purchase behaviour and reviews, quantifying and analysing customer value, analysing customer sentiment in reviews, and combining customer value factors and review sentiment factors in the model. The results of the study show that the model performs best on different indicators, and the area of the main characteristic curve is the largest, which is significantly higher than that of the traditional model. Its hit rate, coverage rate and improvement coefficient also perform well. At the same time, when the sample size increases, the improvement coefficient increases the most, reaching 0.41. In conclusion, the model performs well in customer churn prediction, and it can provide certain reference value for the research field of customer churn prediction.
Hashtags are popular navigability tools in a social media-driven environment. However, social media users have purposely employed a hashtag stuffing strategy, where many unrelated hashtags are added to a post to increase the visibility of the post and drive viewership. The results of the current study suggest a potential negative impact of hashtags on source trustworthiness assessment made by Instagram users through heuristic processing. This research conducted two experimental studies with samples from the overall Instagram population. Study 1 (N = 174) was a 2 × 2 between-subjects factorial experiment designed to demonstrate the positive effects of hashtags' navigability cues on Instagram users' perceived source trustworthiness. Study 2 (N = 185) was a 2 × 2 × 2 experiment that aimed to examine the interactive effect between the visual stimuli of a post and the heuristic cues of hashtags. The current research challenges some of the widely accepted hashtag strategies and provides several practical hashtag usage implications for social media influencers and companies.
In the process of dynamic collaborative mining of user perceived interest points on mobile e-commerce platforms, due to the lack of effective feature classification, the recall rate of interest point data in dynamic collaborative mining of interest points is low. Therefore, a dynamic collaborative mining method for user perceived interest points on mobile e-commerce platforms is proposed. Firstly, coarse grained features of user perceived interest points are initially extracted through clustering algorithms, and their feature values are further extracted using sequence feature extraction algorithms. Then, a user perceived interest prediction model is constructed, and fitting methods are used to achieve feature classification of user perceived interest points. Finally, by designing a dynamic collaborative mining model for user perceived interest points on mobile e-commerce platforms, dynamic collaborative mining is achieved. The experimental results show that the dynamic convergence change of method in this paper interest point data mining is relatively small, and the maximum recall rate is 99%, effectively improving mining performance, thereby providing more accurate and accurate personalised recommendations for mobile e-commerce platforms.
In order to improve the accuracy of cross confidence assessment and shorten the time required for confidence assessment, this article proposes a social media information confidence assessment method based on time series analysis. Firstly, determine the evaluation indicators that affect the credibility of social media information. Then, quantify the evaluation indicators for the credibility of social media information. Finally, a confidence quantitative evaluation function is constructed using time series analysis, and a user information weight allocation matrix is used to configure the weight assignment scheme for each evaluation dimension. By quantitatively calculating the relative importance between various indicators in the comparison criteria layer, the user confidence is finally obtained. The experimental results show that the method proposed in this paper can effectively improve the accuracy and recall of confidence evaluation, with a FI value of 0.9, which verifies the effectiveness of the confidence evaluation method proposed in this paper.
There are problems in personalised recommendation of live streaming e-commerce products, such as low accuracy in user interest mining and weak user relationship strength. Therefore, a personalised recommendation method for live streaming e-commerce products based on multimedia social networks is proposed. First, the user scoring matrix is divided into two interaction matrices by the matrix decomposition method, and the fixed parameter limit matrix dimension is set, and user interest mining is realised by using Euclidean distance calculation. Then, the variance expansion factor is introduced to test the multi-collinearity of the feature, and the contour coefficient is calculated to complete the feature extraction. Finally, user interest and feature data are introduced into multimedia social networks to obtain product feature attention, perform personalised matching, and achieve personalised recommendation. The results show that the method proposed in this paper has good user interest mining performance and strong user relationships.
Traditional social media platforms have low accuracy in identifying false information. Therefore, a method based on multi-modal feature fusion is proposed to recognise false information within social media platforms. This method processes false information data on social media platforms by calculating noise during transmission, and utilises multi-layer management to establish correlations between multi-modal point cloud data. By designing modal grouping and calculating similarity, we integrate information from the three dimensions of time, space, and attributes to supplement the shortcomings of the data. By utilising multi-modal feature fusion algorithms, accurate recognition of false information on social media platforms can be achieved. The experimental results show that using this method can effectively improve the training accuracy of the model and have the ability to resist false data injection attacks, achieving high recognition accuracy.
