基于嵌入表示和加权矩阵分解的线路推荐

Route Recommendation Based on Embedding and Weighted Matrix Factorization

摘要: 为了提升游客体验和增加旅游公司收入，旅游线路推荐成为重要手段。目前已有方法利用了游客与线路的交互历史，提升了推荐性能，但忽略了喜欢相同线路的共现游客对信息、被很多游客喜欢/不喜欢的线路对信息，这些对建模旅客偏好至关重要。因此本文提出了基于嵌入表示和加权矩阵分解旅游线路推荐算法。首先根据游客与线路的交互历史，抽取共现游客对、共现喜欢/不喜欢线路对，建立其词向量模型; 接着利用加权矩阵分解模型对游客与线路的交互历史、共现游客对、喜欢/不喜欢线路对信息进行联合分解完成线路预测；最后在厦航航空公司某旅游公司的真实数据集上评估了算法的有效性。

Abstract: To improve the tourist experience and increase the income of tourism companies, tourism route recommendation has become an important method. Many existing methods train and learn the interaction history between tourists and routes to improve the recommendation performance. However, the user co-occurrence information, codisliked routes and co-liked routes are very important for modeling tourist preference. Therefore, this paper proposes a route recommendation algorithm based on embedding representation and weighted matrix factorization. Firstly, it extracts user’s co-occurrence information, co-disliked routes and co-liked routes based on the interaction history between tourists and routes to establish word vector model by extracting co-occurrence tourists’ likes/dislikes of routes. Then it uses weighted matrix factorization to decompose the interaction history, the co-occurrence traveler information, co-disliked routes and co-liked routes to complete the route recommendation. Finally, the effectiveness of our method is evaluated on a real data set of a travel company of Xiamen Airlines.