大数据电商用户画像及推荐系统

Big Data E-Commerce User Portrait and Recommendation System

互联网的飞速发展使得社会数据量变得越来越庞大。在大数据的背景下,用户对获取自己想要的信息的方式也变得更高了。本文通过搭建电商用户画像推荐系统给用户推荐他们想要的信息。为保证推荐结果的不单一性和推荐结果的新颖性,本文通过基于协同过滤算法,计算用户或物品的相似度,将其作为推荐的依据给用户推荐其想要和新颖的物品。利用Embedding向量的相似性来计算用户和物品之间的Embedding相似度,实现推荐系统的召回,大大地加快了召回效率。同时基于Python使用flask实现推荐系统在线API给某个确定的用户,在他某个使用场景下,推荐他最喜欢的列表。对于冷启动、A/B测试、深度学习技术、难以设计的知识图谱、可交互数据规模有限且奖励信号稀疏、用户画像研究和推荐系统的解释性差这些问题和挑战,本文给出了相应的解决和优化方法来完善推荐系统,使用户更加便捷地获取自己想要的信息并在使用推荐系统时更加满意。

The rapid development of the Internet has made the amount of social data become more and more huge. In the context of big data, users become higher about the way to get the information. This article recommends the information they want to users by building an ecommerce user portrait recommendation system. To ensure the nonsimplicity of the recommendation results and the novelty of the recommendation results, this paper calculates the similarity of the user or items based on the collaborative filtering algorithm. Using the similarity of Embedding vectors to compute the Embedding similarity between users and items, realize the recall of the recommendation system, greatly accelerate the recall efficiency, use flask to implement the recommendation system online API to a certain user based on Python, recommend the favorite list in certain use scene. For the problems and challenges such as cold start, A/B testing, deep learning technology, design knowledge map, limited interactive data scale and sparse reward signal, poor interpretation of user portrait research and recommendation system, this paper gives corresponding solution and optimization methods to improve the recommendation system, then users can easily obtain the information they want and get more satisfactory when using the recommendation system.