Abstract: A heterogeneous information search recommendation algorithm based on multi feature fusion is proposed. The knowledge map technology is used to extract heterogeneous information features, select multi view machine model, learn and integrate local information of multi-feature heterogeneous information by using cooperative attention mechanism, get the importance vector of information through normalization of softmax function, and integrate all local information into the final node, obtain the score function. It uses the corresponding meta path interaction between users and commodities to obtain the global information recommendation optimization objective function of multiple label classification that is combined with the score function and the global information recommendation optimization objective function to obtain the final heterogeneous information search recommendation. The test results show that the algorithm can effectively recommend the information needed by users. The recommended complexity is low, and the normalized discount cumulative gain of searching for heterogeneous information is higher than 0. 35, which has strong recommendation performance and can be applied to solve the problem of information overload.

Key words: multi-feature fusion, heterogeneous information, search recommendation algorithm, user recommendation, multi-view machine model

摘要： 提出基于多特征融合的异质信息搜索推荐算法。利用知识图谱技术提取异质信息特征，选取多视图机模型；利用协同注意力机制学习融合多特征异质信息的局部信息，通过softmax函数归一化处理融合得到信息的重要性向量；利用全部局部信息整合最终节点，获取分值函数；利用用户与商品间相应元路径交互获取多标签分类的全局信息推荐优化目标函数，结合分值函数与全局信息推荐优化目标函数实现异质信息的搜索推荐。算法测试结果表明，采用该算法可有效为用户推荐所需信息，推荐复杂度较低，搜索推荐异质信息的归一化折扣累计增益均高于0.35，具有较强的推荐性能，可应用于解决实际的信息过载问题。

关键词: 多特征融合, 异质信息, 搜索推荐算法, 用户推荐, 多视图机模型