智能系统服务推荐模型研究:基于多任务处理的视角
文献类型：学位论文
| 作者 | 万苓韵 |
| 答辩日期 | 2023-06 |
| 文献子类 | 硕士 |
| 授予单位 | 中国科学院大学 |
| 授予地点 | 中国科学院心理研究所 |
| 其他责任者 | 张警吁 |
| 关键词 | 服务推荐 多任务处理 STOM模型 驾驶分心 |
| 学位名称 | 理学硕士 |
| 学位专业 | 应用心理学 |
| 其他题名 | Modeling of activity recommendation acceptance in intelligent systems: a multitasking-based perspective |
| 中文摘要 | With the development of information technology, today's intelligent assistants can provide proactive activity recommendations for what users could do in the upcoming moments, thereby improving their interaction efficiency. However, activity recommendation, as an emerging technology, still suffers from imperfect recommendation logic and low user acceptance, and there is also a lack of targeted research. The main purpose of this study is to construct a predictive model for the acceptance of activity recommendations from intelligent systems. Given the multi-task management decision essence of the acceptance process, daily cell phone activity recommendations and in-car activity recommendations were selected as typical cases in task-switching and parallel processing contexts, respectively. How the characteristics of the user's ongoing task, those of the task recommended by the system, and other factors such as individual characteristics jointly affect the user's acceptance of the recommendation were examined. Study la first measured the distribution of common activity recommendation scenarios and tasks on key features based on a questionnaire survey (N=220) and selected representative daily cell phone tasks for the follow-up experiment. The hierarchical linear regression model for task evaluation indicated that users' perceived load and involvement of the tasks positively influenced their stickiness in them, and thus might impose restrictions on activity recommendations. Study 1b experimented on daily cell phone activity recommendation acceptance in a task-switching context (N=38), adopting an experimental paradigm that combined task experience and evaluation with simulated switching decisions. Recommendation acceptance was modeled with in-task ECG and EDA physiological indicators incorporated. The hierarchical logistic regression model with subjects at the second level showed that the involvement and priority of the current task negatively predicted the acceptance of the recommended task; the load of the recommended task negatively predicted its acceptance, while its involvement and priority positively predicted it. Physiological indicators such as inter-beat interval and skin conductance response also had additional predictive power for recommendation acceptance. The total explanatory power of the model amounted to 35%. Study 2 further experimented on activity recommendation acceptance in a parallel processing context using an autonomous driving monitoring scenario (N=47), adopting a similar experimental paradigm that combined experience and evaluation of the recommended tasks with simulated responses to recommendations during driving. A takeover task with cognitive judgment requirements was added to the experiment. The acceptance of recommendations and subsequent takeover performance were modeled respectively. The hierarchical logistic regression model for decisions to accept showed that the involvement of the driving task positively predicted the acceptance of the recommended task, while its priority negatively predicted recommendation acceptance; the involvement, priority, and interruptibility of the recommendation task positively predicted its acceptance, while the required completion time and resource competition with the driving task negatively predicted acceptance. In addition, subjects who believed that they could better identify the impact of distraction on driving and regulate their distraction based on driving conditions were more likely to accept recommendations. The total explanatory power of the model was 36%. In terms of safety consequences, the takeover success rate was lower and the reaction time was longer under a higher level of driving task load. Compared to the baseline condition where no recommendation was provided, subjects' takeover performance was impaired after accepting a recommendation task, while their reaction time was shortened after rejecting a recommendation, reflecting the differences in effects recommendation responses. In general, the influence of the main task characteristics on the recommendation acceptance decision has certain commonality in the two scenarios. The recommended task is always more likely to be accepted when the priority of the current task is lower, the load of the recommended task is lower and its involvement and priority are higher. The influence of complementary factors in the parallel processing context further demonstrates the importance of inter-task compatibility. The current study extends the existing findings on multi-task management and driving distraction in the emerging field of activity recommendation. It also provides a rationale for the prediction and improvement of activity recommendation results, and the constraints on their impact. Future research may proceed to explore variants of the model in different application scenarios and task-processing contexts to further promote activity recommendation and enhance its experience. |
| 英文摘要 | 随着信息技术的发展，如今的智能助理能够主动地为用户推荐接下来可进行的活动，从而提升其交互效率。然而，服务推荐作为一项新兴的技术，仍存在推荐逻辑不完善、用户接受度低等问题，也缺乏针对性的研究。本研究的主要目的是构建智能系统服务推荐接受度的预测模型。考虑到其接受过程的多任务管理决策实质，分别选取了日常手机服务推荐、车载服务推荐作为任务切换、任务并行处理语境下的典型场景，考察了用户当前进行中的任务的特征、系统所推荐的任务的特征以及个体特征等其他因素如何共同影响用户对推荐的接受。 研究la首先基于问卷调研伽一220)测量了常见服务推荐场景及推荐任务集在关键特征上的分布，并选取了具有代表性的日常手机任务用于后续实验。基于任务评估的分层线性回归结果表明，用户感知的任务负荷水平及卷入度正向影响他们在任务中的粘性，从而可能为服务推荐带来限制。研究1b在日常手机使用场景中开展了任务切换条件下的服务推荐接受实验伽一38)，采取了先体验并评估任务、再模拟切换决策的实验范式，结合任务中的心电、皮电生理指标，对推荐的接受进行建模预测。以被试为第二水平的分层逻辑回归模型结果表明，当前任务的卷入度、优先性负向影响对推荐任务的接受，而推荐任务的负荷水平负向影响推荐接受，其卷入度和优先性正向影响推荐接受。任务期间的心电RR周期、皮肤电导反应频率等生理指标也对推荐接受具有额外的预测力。模型的总解释力达35%。 研究二在自动驾驶监控场景中开展了任务并行条件下的服务推荐接受实验伽一47)，采取了类似的推荐任务体验及评估、驾驶场景模拟决策相结合的实验范式，并加入具有认知判断要求的接管反应任务，分别对推荐接受决策和后续的驾驶安全绩效进行建模预测。接受决策的分层逻辑回归模型结果表明，驾驶任务的卷入度正向影响推荐接受，其优先性负向影响推荐接受;而推荐任务的卷入度、优先性及可中断性正向影响推荐接受，其所需的完成时长、与驾驶任务的资源竞争程度负向影响推荐接受。此外，认为自身能够较好地判断分心对驾驶的影响、基于驾驶条件调节分心的被试也更倾向于接受推荐。该模型的总解释力达36% o在安全后果方面，较高驾驶任务负荷水平下的接管成功率更低、反应时更慢;相比于未提供推荐的基线条件，被试接受推荐后的接管成功率和反应时都受到损害，而拒绝推荐后的接管反应时反而更快，体现了不同推荐响应情形下的影响差异。 总体而言，主要的任务特征对推荐接受决策的影响在两种场景中具有一定的共性。当前任务的优先性越低，推荐任务的负荷水平越低、卷入度和优先性越高时，推荐任务总是更容易被接受。并行处理语境下补充因素的影响进一步体现了任务间兼容性的重要意义。当前研究在新兴的服务推荐领域对已有的多任务管理、驾驶分心研究结论进行了补充，并为服务推荐效果的预测和改善、其影响的约束提供了依据。未来的研究可以继续探索不同应用场景和任务处理形式中的模型变体，以进一步推广并提升服务推荐的使用体验。 |
| 语种 | 中文 |
| 源URL | [http://ir.psych.ac.cn/handle/311026/46096]  |
| 专题 | 心理研究所_社会与工程心理学研究室 |
| 推荐引用方式 GB/T 7714 | 万苓韵. 智能系统服务推荐模型研究:基于多任务处理的视角[D]. 中国科学院心理研究所. 中国科学院大学. 2023. |
入库方式： OAI收割
来源：心理研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。