融合用户、上下文和系统：基于量子理论的信息检索模型研究

The user's information seeking (IS) process is far beyond a classical information retrieval (IR) paradigm of a system simply matching queries and documents. Instead, IS is highly contextual and interactive in nature, particularly in the emerging Web2.0 social media environment. Users often seek information through interactive exploration over a specific search domain for learning, discovery, decision making and completing certain complex tasks. Thus it is our long term version that the future online IS will be converging with user cognition, decision making and problem solving. IS involves a number of inter-related key aspects, including the handling complex information objects, complex information needs, the user's interaction both with the system and with other users, the context, the user's cognitive state, and decision making (e.g. which particular information items are relevant and in what sense they are relevant). These aspects are dynamic, evolving and interacting with each other in complex and unpredictable ways. The classical IR paradigm and corresponding conventional search technologies do not satisfactorily deal with these aspects and their interactions in an integrated and holistic manner. To make a genuine breakthrough requires a radically new way of thinking and synergy of expertise across various relevant disciplines including information retrieval modelling, user modelling, cognition and decision making. ..This project aims to provide a non-classical paradigm for unifying these key aspects and for comprising seeking, interaction and decision making in the complex information environments of the future. The theoretical inspiration derives from the recent emerging quantum-like theories addressing problems in cognition, artificial intelligence and organisational decision-making using quantum theory (QT) formalisms. QT seems to fit naturally with how people link concepts together, evaluate probabilities and make decision. To this end, we will develop QT-inspired poly-representation of complex information objects, inforamtion needs and multidimensional relevance, context detectors (quantum operators acting on an information space), and decision functions (based on quantum probability, superposition and interference), which integrate into the paradigm. An initial prototype system will be implemented and tested to utilise the paradigm to underpin adaptive and personalised information seeking in a Web2.0 social media environment. The resulting prototype system will be systematically evaluated a series of transition-based simulated user and real user evaluations.

本项目将研究基于量子理论框架的新型信息检索模型和方法论，用以融合复杂信息对象的多元表达，复杂信息需求和多维度相关性的刻画，以及上下文情境，用户的认知状态和决策生成（例如，判定哪些信息是相关的以及在何种意义上是相关的）。我们的理论灵感源于认知科学、人工智能和决策支持等相关领域最近迅速发展的"类量子"理论模型，尤其是其对观察者的中心位置的强调以及对心智与实在之间交互关系的建模，可自然解释人们进行概念联想、评价可能性（主观概率）和产生决策的思维过程。我们将发展有效整合于量子范式下的复杂信息对象、复杂信息需求和多维度相关性的多元表示，动态上下文多元表达和演化模型和决策函数（基于量子概率、量子叠加和干涉）。在此基础上，我们将实现一个初步原型系统，测试量子范式作为未来以用户为中心的自适应性信息获取系统的建模原则的可行性。最终的系统有效性将通过一系列基于状态迁移的仿真用户实验和真实用户满意度评价来验证。

本项目研究基于量子理论框架的新型信息检索模型和方法论,用以融合复杂信息对象的多元表达,复杂信息需求和多维度相关性的刻画,以及上下文情境,用户的认知状态和决策生成(例如,判定哪些信息是相关的以及在何种意义上是相关的)。我们的理论 灵感源于认知科学、人工智能和决策支持等相关领域最近迅速发展的“类量子”理论模型, 尤其是其对观察者的中心位置的强调以及对心智与实在之间交互关系的建模,可自然解释人们进行概念联想、评价可能性(主观概率)和产生决策的思维过程。本项目的研究重点是：首先，基于希尔伯特空间、张量积、量子干涉和量子纠缠等量子元素实现了用户信息需求、信息对象的多元子空间表示；其次，基于密度矩阵、有算子和量子干涉实现了用户在探索式信息检索过程中信息需求的动态演化。项目具体研究内容可以概括为以下几个方面：1) 对个性化搜索和探索式搜索两种新一代互联网信息搜索技术的基础理论和关键技术研究；2) 基于用户认知状态和认知水平的信息检索; 3) 基于偏差-方差统计分析的的信息检索建模、分析和评测框架; 4) 基于玻尔兹曼机的模型选择理论以及相关信息检索模型；5) 量子信息检索和量子认知。围绕这些研究内容，在本项目资助下获得大量研究成果：SCI一区或CCF推荐A类期刊（ACM TOIS，ACM TIST）论文2篇；被接收SCI二区或CCF推荐B类期刊（IPM, PRA, Quantum Information Processing, JASIST）论文6篇；发表（或被接收）SCI三区或CCF推荐C类期刊论文7篇；SCI四区2篇；Science增刊论文1篇；CCF推荐A类会议论文（SIGIR，IJCAI, WWW等）7篇；CCF推荐B类会议论文（CIKM，EMNLP, BIBM等）4篇；CCF推荐C类会议论文2篇；信息检索领域重要国际会议论文（EI检索）11篇；SIGIR会议Work Shop论文2篇；中文核心期刊论文4篇；国家软件著作权4件；申请国家专利2件。这些成果中有一些在同行范围内引起了较大的反响。有关量子语言模型的研究在世界范围内取得了领先地位。