地理信息检索中语境的获取、推理及应用

TThe aim of geographical information retrieval (GIR) is to find target objects based on their relations to reference objects. Therefore, correctly understanding the real meanings of target object, relation and reference object in a query sentence is very important in a GIR process. Since there are a number of distinct categories of place expressions, and the spatial relation terms employed in natural language are highly ambiguous, the query sentences for geographical information retrieval are usually vague and ambiguous. Web search engines generally treat the queries without considering the user, or the context in which the user made the quest, so the results for given queries sometimes are not what users expect. The project proposes we should consider context in GIR process for the purpose of disambiguation. There are two kinds of contexts revolved in GIR. One is representation context, which are the words in a query sentence that can characterize the real meanings of the three elements used in GIR. The other is user context, such as the location and purpose of the user, and other individual characters of the user, which can help us to fix the real meanings of the three elements. The project studies the methods and key techniques of acquiring these two kinds of context information, uncertainty reasoning with contexts and applying contexts in GIR. The research includes three parts: 1) Cognition models of contexts. Comprehensively and synthetically analyze all kinds of context involved in GIR and their relations; acquire the cognition model of each kind of context by methods of web data mining or human subject tests; and then represent cognition models with fuzzy mathematical method. 2) Context formalization and reasoning. Build context ontology; formalize contexts with first order logic and study on context transition, change of focus, context sharing; construct a Bayesian network based on context ontology and fuzzy cognition models; and disambiguate query sentences with Bayesian probability reasoning. 3) A prototype of context based geo-spatial search engine. Apply context information in process of GIR and implement a prototype of context based geo-spatial search engine; resolve the method of result ranking by the analytic hierarchy process; and evaluate the search result. By the study of this project, we hope it can disambiguate the query sentence in GIR by using context information, so as to improve the veracity of query results, to make the results be more qualified with human's cognition and meet the needs of users.

在地理信息检索中，由于文本中自然语言表达的语义多样性，地名表达和空间关系的不确定性，以及不同用户认知上的差异，往往造成对查询语句理解上的歧义，导致检索结果与用户的期望相去甚远。本项目通过研究地理信息检索中相关语境的获取和推理等关键问题，将语境用于地理信息检索。首先，对地理信息检索中涉及的语境进行深入分析，综合考虑检索语句中所蕴含的语境信息，以及用户的查询目的、用户所处的地理位置、个人偏好等用户语境，通过认知实验和网络数据挖掘，获取与地理信息检索有关的各种语境的认知模型；在此基础上，研究语境的形式化表达和不确定推理方法，解决不确定检索结果的匹配度和排序等关键技术；研究将多方面的语境信息综合用于地理信息检索的方法，以个性化餐厅推荐系统为例，实现基于语境的地理信息搜索原型系统。通过本项目的研究，达到消除检索语句歧义的目的，提高地理信息检索的准确度，得到符合大众认知、用户满意的检索结果。

互联网的快速发展为大众提供了海量信息，为了有效利用信息，需要使人们从网络中更快获取所需要的信息。用户对地理服务的选取受到个体特征、行为的时空语境以及社交好友的影响，因此考虑用户个人的兴趣偏好和时空语境的信息检索有助于提高用户对检索结果的满意度。项目提出将行为语境和社交语境用于地理信息的地理信息的推荐，围绕语境认知模型获取、语境的形式化表达和推理展开，首先研究了利用大数据获取行为语境和社交语境的数据挖掘方法；进而研究了基于语境的个性化地理信息推荐方法，提出空间异构信息网、特征空间分层的矩阵分解的语境表达和计算模型；以语境模型为基础，实现了基于空间异构信息网中语义路径相似性的个性化推荐、基于矩阵分解的POI推荐、空间级联模式下的个性化推荐、群体偏好增强的群推荐等多种地理信息推荐方法，并实现了推荐系统原型和推荐结果可视化。项目在大数据挖掘、语境知识表达和计算方面具有科学意义，将多种语境信息融于地理信息服务，转被动搜索查询为主动推荐，从而丰富和便利用户获取地理信息的手段，在移动互联网和大数据应用方面具有应用前景。