基于知识图谱和图注意力的众包任务推荐算法

为解决目前众包任务推荐存在未考虑任务文本信息和数据稀疏的问题，提出一种基于知识图谱与图注意力的众包任务推荐模型。该模型首先利用自然语言处理技术提取任务文本信息中的关键要素，用于丰富图谱信息和缓解数据稀疏性；通过融合用户—任务交互图中的协同信息来构建协同知识图谱，在协同知识图谱中按协同邻居的类型分别运用图注意力网络；为获取用户准确的偏好，聚合邻居信息时按注意力得分从高到低采样固定数目的邻居；最后通过聚合不同类型的协同信息生成用户和任务的嵌入表示并得到交互概率。在构建的众包数据集上进行实验的结果表明，该模型在AUC、精准率、召回率和NDCG四个指标上均优于基线模型，验证了模型的可行性和有效性。

Crowdsourced task recommendation algorithm based on knowledge graph and graph attention network

In order to solve the current problems of crowdsourcing task recommendation that don''t consider task text information and data sparsity, this paper proposed a crowdsourcing task recommendation model based on knowledge graph and graph attention network. The model first used natural language processing techniques to extract key elements from task text information for enriching graph information and alleviating data sparsity. Then it constructed a collaborative knowledge graph by fusing the collaborative information in the user-task interaction graph, and applied the graph attention network in the collaborative knowledge graph according to the types of collaborative neighbors. To obtain accurate user preferences, it sampled a fixed number of neighbors from highest to lowest attention score when aggregating neighbor information. Finally, it aggregated different types of collaborative information to generate embedding representations of users and tasks, and obtained interaction probabilities. Experiment on the constructed crowdsourcing dataset shows that the model is superior to the baseline model in four metrics: AUC, precision, recall and NDCG, which verifies the feasibility and effectiveness of the model.