The growing availability of data in electronic form, the expansion of the World Wide Web (WWW) and the accessibility of computational methods for large-scale data processing have allowed researchers in Information Retrieval (IR) to design systems which can effectively and efficiently constrain search within the boundaries given by context, thus transforming classical search into contextual search. Because of the constraints imposed by context, contextual search better focuses on the user's relevance and improves retrieval performance, since the out-of-context aspects of the search carried out by users that are likely linked to irrelevant documents are left apart.

This survey introduces contextual search within a computational framework based on contextual variables, contextual factors and statistical models. The framework adopted in this survey considers the data observable from the real world entities participating in contextual search and classifies them as whatwe call contextual variables. The contextual variables considered are content, geotemporal, interaction, and social variables. Moreover, we distinguish between contextual variables and contextual factor: the former is what can be observed, the latter is what cannot be observed, yet this is the factor affecting the user's relevance assessment. Therefore, in this survey, we describe how statistical models can process contextual variables to infer the contextual factors underlying the current search context.

In this survey we provide a background to the subject by: placing it among other surveys on relevance, interaction, context, and behavior; providing the description of the contextual variables used for implementing the statistical models which represent and predict relevance and contextual factors; citing and surveying useful publications to the reader for further examination; providing an overview of the evaluation methodologies and findings relevant to this subject; and briefly describing some implementations of contextual search tools.

电子形式的数据的可用性不断增长，万维网（WWW）的扩展以及大规模数据处理的计算方法的可访问性，使得信息检索（IR）的研究人员可以设计出可以有效地约束搜索的系统在上下文给定的范围内，从而将经典搜索转换为上下文搜索。由于上下文所施加的约束，上下文搜索更好地集中于用户的相关性并提高了检索性能，因为将可能链接到不相关文档的用户进行的搜索的上下文外方面分开了。

该调查在基于上下文变量，上下文因素和统计模型的计算框架内引入了上下文搜索。本调查采用的框架考虑了可以从参与上下文搜索的现实世界实体中观察到的数据，并将其归类为我们称为上下文变量。考虑的上下文变量是内容，时空，交互和社会变量。此外，我们区分了上下文变量和上下文因素：前者是可以观察到的，后者是无法观察到的，但这是影响用户相关性评估的因素。因此，在本次调查中，我们描述了统计模型如何处理上下文变量以推断当前搜索上下文所基于的上下文因素。

在本次调查中，我们通过以下方式为主题提供了背景：将其置于相关性，交互性，上下文和行为的其他调查之中；提供用于实施统计模型的上下文变量的描述，这些变量表示并预测相关性和上下文因素；引用和调查有用的出版物，以供读者进一步检查；提供与该主题相关的评估方法和发现的概述；并简要描述上下文搜索工具的一些实现。