在线短租平台双向信任计算和匹配机制研究

Both buyers and sellers of online short-term rental platforms are sensitive to personal and property safety, but trust assessment information is extremely limited, so that it is difficult to establish bilateral trust between buyers and sellers and resulting in low matching rate, and it becomes a dilemma for online short-term rental platforms and an academic problem. Our research integrates multidisciplinary theories and aims to make the following contributions: 1. To cope with limited information limitations, new trust dimensions are expanded using the new trust theory of "vision trust", and deep learning techniques are used to map unstructured data to trust, such as images; 2. Considering privacy, a two-stage of feature selection methods is designed. In the first phase, a rough set feature selection algorithm integrated with an evolutionary algorithm obtains low-dimensional feature sets that approximating the decision-making ability of all feature, in the second phase restricted-random feature sub-spaces are constructed using heuristic methods, then optional low-dimensional feature sets are obtained, which reduces the needed evaluation information; 3. Evaluation models using feature subspace and evidence theory are Established, multiple evaluations are made using multiple feature combinations, and an integrated framework are designed to integrates multiple evaluation results, which can make full use of incomplete data to make effective evaluation, and break the limitation of existing models using fixed features for evaluation. 4. First , a two-aspect method that integrates the explicit and implicit modes is designed to predict bilateral trust , and the matching mechanism is designed by maximizing the potential benefits of buyers and sellers to matching honest buyers and sellers. This research can provide new research perspectives and ideas for on-line trust management for online short-term rental platforms, and can be further extended to other areas of the sharing economy.

在线短租平台买卖双方人身财物安全敏感，但信任评估信息极为有限，双向信任难建立导致匹配率低下，成为在线短租平台的现实困境和学术难题。本项目融合多学科理论，拟形成以下创新:①基于“视觉信任”等理论拓展新维度，以机器学习方法完成图像等非结构化数据向信任映射，应对有限信息局限；②考虑隐私问题，设计二阶段特征选择方法，第一阶段粗糙集特征选择算法融合进化算法获得逼近全体特征决策能力的特征组合，第二阶段基于特征组合启发式构建有限随机特征子空间，提供多种可选低维特征组合；③以特征子空间和证据理论构建评估模型，用多种特征组合多次评估，设计集成框架融合多次评估结果，突破现有模型使用固定特征组合的局限，可在不完备数据下有效评估；④双向信任感知的匹配机制先融合显式和隐式模式预测双向信任，再以买卖双方潜在收益最大化为目标进行匹配。研究成果可为在线短租平台信任管理提供新视角和思路，并可推广到共享经济其它领域。

在线短租作为共享经济代表性行业之一，备受学术和商业各界关注。区别于传统电商交易，在线短租交易由陌生的买卖双方在第三方在线短租平台下完成，涉及线上交易和线下交互。信任对在线短租至关重要。我国信用体制尚待建设和完善情景下，对在线短租信任进行研究，对维护交易环境和促进经济健康、可持续发展具重要意义。本研究内容包括：基于多模态数据的信任特征提取、核心信任特征识别、信任预测和匹配模型。.(1) 基于多模态数据的信任特征提取。基于“视觉信任”理论，基于图像分析技术，分析房东头像，提取特征包括人脸数量、视觉年龄、颜值等；分析房屋图像，提取特征包括：实拍、户外和室内等；基于文本挖掘技术，分析房屋介绍文本，提取特征包括：词性数量、文本长度和文本可读性；分析评论，提取特征包括：情感极性、文本信息质量等。.(2) 融合粗糙集理论和智能算法的特征选择算法设计。提出嵌套等价类概念，提取决策表中知识，不仅压缩数据，更可有效监督特征选择过程并加速计算。融合智能算法，基于嵌套等价计算适应度，可突破性能瓶颈，快速形成特征多个子空间，可有效缓解隐私问题，提供可选低维特征组合。.(3) 设计信任评估和预测模型。考虑核心信任特征随时间动态变化、数据缺失等情况。构建在线短租信任计算框架，通过特征选择持续更新信任特征子集，可跟踪核心信任特征动态变化；整合多个低维信任特征子集，评估和预测房东感知信任，可缓解数据缺失导致模型失效的问题。.(4)信任感知的匹配模型构建。分析卖家 (MGC)和买家生成内容(UGC)，.识别双方关注点并完成其在关注点上投影计算，形成买卖方嵌入向量，从而计算匹配度。考虑信任动态性，设计基于矩阵分解的时间感知信任匹配规则挖掘模型，获得带全记忆、持续遗忘和选择性遗忘的信匹配规则。.分析结果表明，多模态数据，包括图像、文本和数值，均对卖家感知信任计算具有重要影响，安全性、功能性、真实性相关内容、房东信誉形象、文本质量等信任特征对买卖双方间信任构建和匹配具有关键作用。