短视频推荐系统的设计与实现.docxVIP

短视频推荐系统的设计与实现 摘要 在互联网行业中，几乎所有领域都有应用推荐系统。例如我们日常接触到的美团、今日头条、腾讯视频等应用，都具有强大的推荐系统来进行个性化推荐的服务。从数学层面来看推荐算法，推荐算法的实际上是一种关于项目满意度的拟合函数。推荐算法的研究对象主要包括物品特征以及用户特征。这些维度丰富的特征通常用于模型训练，以此进行特征工程。在特征工程中构建数据集，并且用使用合理的学习监督算法来训练，可以获得一个模型，最后实现推荐符合用户偏好的内容。在短视频行业中，推荐系统的任务就是筛选出用户喜爱的短视频。 可以选择的推荐算法很多，如协同过滤，逻辑斯蒂回归，基于DNN的模型，FM等。本系统使用的方式是协同过滤，基于余弦相似度计算进行召回，然后产生推荐结果。本设计首先以moment in time数据集作为数据源。接着利用Python脚本模拟用户对这些带标签的短视频进行评分生成数据表。然后，使用django框架搭建后台和前端系统。最后，一系列评价指标对本推荐系统进行评测，并尝试从结果分析找到优化短视频推荐效果的新思路，新方法。 关键词：推荐系统，推荐算法，协同过滤 Abstract In the Internet industry, recommendation systems are used in almost all areas. Among the software we use frequently like Meituan, Todays headlines, Tencent video and other applications, all of them have a strong recommendation system for the services about personalized recommendation. In a mathematical perspective, the recommendation algorithm is actually a fitting function about the satisfaction of the items. The research object of the recommendation algorithm mainly includes the characteristics of the items and the characteristics of the users. These dimension-rich features are often used for model training to perform feature engineering. The data set is then built and trained using a reasonable learning supervision algorithm. Finally, get a model and use this model to recommend what users might like. In the short video industry, the task of the recommendation system is to screen out short videos that users may like. There are many recommended algorithms to choose from, such as collaborative filtering, logistic regression, DNN-based models, FM, etc. In this system, the way to be used is collaborative filtering. First calculate the cosine similarity, then recall these items, and finally generate the recommended results. In this design, the moment in time data set is used as the data source. The Python script is then used to simulate the users scoring of these tagged short videos. Then, build the backend and frontend systems using the django framework. The next step is to