Understanding WeChat User Preferences and “Wow” Diffusion

Abstract

WeChat is the largest social instant messaging platform in China, with 1.1 billion monthly active users. "Top Stories" is a novel friend-enhanced recommendation engine in WeChat, in which users can read articles based on preferences of both their own and their friends. Specifically, when a user reads an article by opening it, the "click" behavior is private. Moreover, if the user clicks the "wow" button, (only) her/his direct connections will be aware of this action/preference. Based on the unique WeChat data, we aim to understand user preferences and "wow" diffusion in Top Stories at different levels. We have made some interesting discoveries. For instance, the "wow" probability of one user is negatively correlated with the number of connected components that are formed by her/his active friends, but the click probability is the opposite. We further study to what extent users' "wow" and click behavior can be predicted from their social connections. To address this problem, we present a hierarchical graph representation learning based model DiffuseGNN, which is capable of capturing the structure-based social observations discovered above. Our experiments show that the proposed method can significantly improve the prediction performance compared with alternative methods.

Social Theory Aware Influence Maximization Approach for Target Community Detection

“…[19] studied the community search problem on directed graphs, given the directed graph and query nodes, based on the minimum in/out-degree of nodes, looking for a densely connected subgraph containing query nodes; Literature [20]. A two-stage local community detection method was designed, using breadth-first extension to locate local communities by seed replacement and expansion; Literature [21] studied the use of personalized PageRank model to identify the community where a group of seed nodes reside, which relied on seed Select and assume that there are some benchmark communities, but cannot efficiently search for specific target communities in large graphs based on query requests; Literature [22] proposed a community detection method based on multi-query random walks, which considers The interactive information Social Theory Aware Influence Maximization Approach for Target Community Detecti between different walkers accurately captures the community; Literature [23] used the community affiliation of the seed nodes as prior information, marked other seed nodes by color, and introduced a coloring-based random walk mechanism to separate the seed nodes The color is propagated to the rest of the nodes in the graph, and the communities where the seeds of different colors are located are mined. However, the above methods apply to simple charts and only consider the direct interaction information between nodes, ignoring the influence of high-order neighbours on the importance of nodes, so they are not effective enough for user interest discovery.…”

“…However, the above methods apply to simple charts and only consider the direct interaction information between nodes, ignoring the influence of high-order neighbours on the importance of nodes, so they are not effective enough for user interest discovery. In these methods, the local community structure is usually captured by metrics such as local modularity, k-cluster, k-truss and k-core [23]. In recent years, many novel network models have emerged, and community search methods have also carried out some preliminary research on these diverse types of networks, including heterogeneous networks, public-private network, geo-social networks and hierarchical attribute graphs and so on.…”

Multimodal Dialog Systems with Dual Knowledge-enhanced Generative Pretrained Language Model

“…As an emerging technique, pretrained language models have been arresting much research attention [10], [11], [12], [24], [25] and achieve remarkable success in plenty of NLP tasks. Initially, Word2vec [26] and GloVe [27] are proposed to obtain pretrained word embeddings based on shallow architectures.…”

Energy Demand Forecast in Yunnan Province Based on Seq2Seq Model

“…The trend extrapolation method is mainly based on past and current development trends to infer possible future states, such as time series method [1], energy elasticity coefficient method [2], and input-output method [3][4][5]. The scenario analysis method mainly identifies the external factors affecting the development of the main body of the research through the study of the environment and simulates various cross scenarios that may occur in the external factors to analyze and predict various possible prospects [6][7][8][9].…”