大数据环境下，科技文献数量以惊人的速度增长。在检索相关文献时，人们普遍面临着知识迷航和信息过载问题。在此背景下，学术搜索系统针对用户特征开展个性化服务逐渐引起学术领域的关注。用户画像作为用户研究和信息技术相结合的产物，以标签形式抽象出用户特征有助于迅速了解用户。 相比电子商务平台，构建学术搜索系统用户画像难点在于：（1）由于允许匿名访问和共享访问，难以根据日志判定哪些行为来自特定个体。个体用户边界判定会影响特定用户相关数据的收集，进而影响用户画像的精准性。（2）用户画像的核心在于用户标签设计和提取。目前，国内外研究还未针对学术搜索系统中的用户建立用户标签体系。此外，用户画像表示和更新也是研究难点。（3）目前，国内研究大多探讨用户画像模型的设计，很少基于真实数据利用相关技术进行实证研究。另外，由于没有公认的评测数据集和评测指标，用户画像评测困难。 为了帮助学术搜索系统更好开展个性化服务，本文基于用户搜索日志对学术搜索系统中用户画像进行研究，主要围绕以下方面开展：（1）用户画像前期——个体数据融合。由于用户画像是基于用户真实数据建立的用户模型，获取特定个体的相关数据是进行画像的必要工作。为了解决学术搜索系统用户日志数据中的个体用户行为关联问题，本文提出基于科研用户小数据和随机森林分类的个体用户识别方法，对学术搜索系统中个体用户进行识别。（2）用户画像中期——用户画像构建。在获得特定个体的相关数据后，本文从人口属性标签、资源属性标签、行为属性标签和主题兴趣标签设计用户标签体系，综合考虑多种方式，实现用户标签的提取和评价。介绍并实现了基于BOW、LDA和Word2Vec的用户标签表示，使用训练好的Word2Vec模型对用户兴趣标签进行语义扩展，实现对用户兴趣标签的预测。（3）用户画像后期——个性化信息服务。用户画像的目标是通过精准刻画用户，从而提供个性化服务。本文对用户画像如何提升学术搜索系统个性化服务进行探讨，主要围绕个性化搜索、资源推荐和消息推送三个方面展开论述，目的是为了让用户画像真正落实到科技文献服务工作中。

In big data environment, the number of scientific literature has grown at an alarming rate. When searching for relevant literature, people are generally faced with the problem of knowledge voyage and information overload. Therefore, providing personalized service for user has become more and more popular gradually in academic search system. User portraits，as product of user research and information technology, abstract user tags from user characteristics，can help people quickly understand users. Compared with the e-commerce platform, the difficulties to construct user portrait in academic search system lie in: (1) When users access the system without logging in or many users use the same account，it is difficult to determine specific individual behaviors from the whole log data. That may affect the accuracy of the user portrait. (2) The key of the user's portrait is to design and extract user tags from user characteristics. At present, there has not established a special user tags structure for users in academic search systems among many studies in domestic and foreign. In addition, representation and update of user portrait are also research difficulties. (3) At present, most researches explore the design of user portrait, and rarely conduct empirical research based on real data. Additionally, due to the lack of recognized evaluation data sets and evaluation indicators，it is difficult to evaluate the precision of user portrait. In order to help the academic search system to provide more personalized service, we focus on user portrait in academic search system based on the user search log. The main contents are as follows： (1) During the prophase of user portrait, we mainly integrate the log data of individual users. It is a necessary work to build user portrait for which user portrait is a user model based on real data of the user. It is difficult to establish user behaviors association from user log data in academic search system. Based on scientific research user small data and random forest classification, we propose individual user identification methods in academic search system. (2) During the metaphase of user portrait, we build user portrait of the academic search system. After obtaining the relevant data of individuals, we design user tags structure, which mainly consists of population attribute tags, resource attribute tags, behavior attribute tags, and topic interest tags for academic search system. Then，we extract user tags combining many methods and evaluate the precision of tags. Additionally, we introduce and realize the representation of user tag, based on BOW, LDA and Word2Vec respectively. Finally, we semantically extend the user interest tags to predict the future interest tags by training Word2Vec model. (3) During the anaphase of user portrait, we mainly discuss the application of user portrait in personalized information services. Aimed to make user portrait truly implemented in the scientific literature service，we discuss how to use the user portrait to enhance the personalized service of the academic search system. We mainly focusing on three aspects: personalized search, resource recommendation and message push.