联系我们

论文查询结果	返回搜索

论文编号：	15596
作者编号：	2320234198
上传时间：	2025/12/7 11:16:08
中文题目：	基于用户体验的对话式搜索引擎服务评价研究
英文题目：	Research on User Experience-Based Evaluation of Conversational Search Engine Services
指导老师：	姚欣林
中文关键字：	对话式搜索引擎；用户体验；LDA主题分析；主成分分析
英文关键字：	Conversational Search Engines; User Experience; LDA Topic Modeling; Principal Component Analysis
中文摘要：	人工智能的迅猛发展正深刻变革信息检索环境，推动其向智能化演进，其中传统搜索引擎向对话式搜索引擎的转变尤为显著。尽管传统搜索引擎评价体系已相对成熟，但现有对话式搜索引擎评价研究多聚焦于用户信息查询环节，忽视了信息选择与利用行为的复杂性与多维性。为弥补这一研究缺口，本研究旨在以用户体验为核心，在系统梳理传统搜索引擎与问答社区评价维度的基础上，构建一个综合评价体系。 基于上述目标，研究首先通过文献调研构建初步指标体系，进而以5076条微博用户内容为语料，采用LDA主题模型提炼出“情感交互深度”这一用户隐性需求指标，实现了文献指标与真实用户需求的交叉验证与补充。随后，研究设计Likert五级量表问卷，回收有效样本303份，通过探索性因子分析与主成分分析迭代优化，将34个初始指标精炼为18个核心指标，并客观赋权，创新性地提出“交互进化能力—答案生成质量—信息控制质量—系统运行性能—情感计算能力”五维整合框架，实现了从“系统中心”向“体验-系统-内容”三元协同视角的范式转换。为验证模型实用性，本研究选取豆包、ChatGPT、DeepSeek三个主流平台进行实证研究，发现三者在交互进化能力、答案生成质量、信息控制质量、系统运行性能及情感计算能力五维度上存在显著差异：ChatGPT在交互进化与信息控制方面表现最优，豆包在移动端体验与搜索效率上具有优势，DeepSeek则在中文资源聚合与信息广度上领先，但三者普遍存在情感计算能力不足与系统稳定性瓶颈。 本研究不仅为对话式搜索引擎服务优化提供了差异化改进路径，也为后续研究与实践提供了可移植的理论工具与实证参照，推动了对话式搜索引擎评价研究从“技术视角”向“用户体验全链路视角”的深化发展。
英文摘要：	The rapid advancement of artificial intelligence is profoundly transforming the information retrieval landscape toward intelligent systems, with a key manifestation being the shift from traditional search to conversational search. While evaluation frameworks for traditional search engines are relatively mature, existing research on conversational search engines predominantly concentrates on the information-seeking phase, neglecting the complexity and multidimensionality of users’ information selection and utilization behaviors. To bridge this gap, this study centers on user experience and proposes an integrated evaluation framework for conversational search engines that encompasses the entire user information behavior chain (querying, selecting, utilizing). By systematically reviewing the evaluation paradigms of traditional search engines and Q&A communities, this study innovatively introduces a five-dimensional integrated framework—“interaction evolution capability, answer generation quality, information control quality, system operational performance, and affective computing capability”—realizing a paradigm shift from a “system-centric” to a “user-experience–system–content” triadic perspective. To achieve this objective, the research first constructs a preliminary indicator system through a systematic literature review. Subsequently, employing Latent Dirichlet Allocation (LDA) topic modeling on 5,076 authentic user-generated comments from Weibo, the study uncovers users’ latent demand for “affective interaction depth,” thereby cross-validating and enriching the literature-based indicators. A five-point Likert-scale questionnaire was designed and distributed, yielding 303 valid responses. Through iterative exploratory factor analysis (EFA) and principal component analysis (PCA), the initial 34 indicators were refined into a scientifically grounded set of 18 core indicators with objective weights. To validate the model’s practicality, an empirical study was conducted across three mainstream platforms: Doubao, ChatGPT, and DeepSeek. The empirical findings reveal significant performance variations among platforms across the five dimensions. ChatGPT excels in interaction evolution capability and information control quality; Doubao demonstrates advantages in mobile experience and search efficiency; while DeepSeek leads in Chinese resource aggregation and information breadth. However, all three platforms exhibit common bottlenecks in affective computing capability and system stability. This research not only provides differentiated improvement pathways for service optimization but also offers transferable theoretical instruments and empirical references for future research and practice, advancing the evaluation studies of conversational search engines from a “technical perspective” toward a “full-chain user-experience perspective”
查看全文：	预览  下载(下载需要进行登录)