| 摘要: | 群眾感知（Crowdsensing）已成為一種重要的大規模資料收集方法，該方法利用用戶的行動裝置(如智慧型手機)內建的感測器來收集資料。用戶透過參與群眾感知任務並提交感測資料來獲取獎勵，群眾感知平台則依據收集到的資料或資訊，並提供以人為本的服務。然而群眾感知系統中面臨資料品質不可靠的挑戰，系統中存在惡意用戶可能會為了獲取獎勵而提交虛假的資料，此種惡意行為會導致資料不準確，進而影響整體服務品質，因此需要一套穩健的激勵機制。
現有研究激勵機制大多並未考慮用戶之間的社交關係與資料新鮮度，因此本研究提出一種結合資料新鮮度與拍賣理論的社群群眾感知激勵機制。在激勵機制環境層面，我們參考社會心理學，將社群群眾感知平台中的用戶分為誠實和惡意兩種類型用戶，不同類型的用戶會展現出不同的行為特徵與策略。在激勵機制架構層面，本研究設計一套以反向拍賣為核心的激勵機制，並整合資料真實性發現、社群信任、社群影響及個人信譽等模型，進一步提出用戶間感知信任模型，用以描述用戶在社群導向群眾感知系統中的信任關係。在激勵機制運作層面，先以個人信譽與社群信任分數篩選並排除不可信用戶，再於預算限制內綜合歷史任務貢獻與競標資訊，利用反向拍賣選出適合的用戶，用戶完成任務後，系統透過歷史與即時資料相互比對，依據用戶任務貢獻與表現，動態調整其歷史行為紀錄，並透過用戶彼此監督與互相激勵來提升感測資料品質。
本研究以真實社群網路資料為基礎進行實驗，並根據用戶規模及惡意用戶比例設計多種實驗情境，以評估機制在不同環境下的表現。結果顯示，在惡意用戶占比為20%的情境下，所提出機制之系統資料準確率明顯高於僅採用信譽機制的26.88%，亦優於傳統機制的38.41%，同時在預算限制下仍能有效排除惡意用戶，確保所蒐集資料之可靠性與新鮮度。;Crowdsensing has emerged as an important paradigm for large-scale data collection. Sensing data can be gathered by leveraging the built-in sensors of users’ mobile devices (e.g., smartphones). Users contribute sensing data by participating in crowdsensing tasks in exchange for rewards, while crowdsensing platforms utilize the collected data to provide human-centric services. However, crowdsensing systems face challenges caused by unreliable data quality. Malicious users may intentionally submit falsified data to fraudulently obtain rewards, leading to inaccurate sensing outcomes and degrading overall service quality. Therefore, a robust incentive mechanism is required to ensure system reliability.
Most existing incentive mechanisms in crowdsensing research do not consider social relationships among users or data freshness. To address this, this study proposes a data freshness and reverse auction–based incentive mechanism for a social crowdsensing platform. In this environment, users are classified into honest and malicious types inspired by social psychology, with each type demonstrating distinct behavioral characteristics and strategies. The architecture of the proposed mechanism integrates an incentive framework based on a reverse auction, which incorporates a data truth discovery mechanism, a social trustworthiness model, a social influence model, and a personal reputation model. Furthermore, a sensing trust model is proposed to describe trust relationships among users within the platform. Based on this architecture, the system first excludes untrustworthy users based on their personal reputation and social trustworthiness scores. Then, under a budget constraint, the system selects suitable participants based on auction theory. After users complete a crowdsensing task, the platform compares the sensing data with historical data and dynamically updates users′ historical behavior records based on their contribution.
The simulation environment in this study is based on a real-world social network dataset. Multiple simulation scenarios are designed by varying the number of users and the ratio of malicious users to evaluate the performance of the proposed mechanism. Simulation results show that when the ratio of malicious users reaches 20%, the proposed mechanism demonstrates significantly higher data accuracy, outperforming the reputation-based mechanism by approximately 26.88% and traditional mechanisms by 38.41%. Moreover, the mechanism can effectively exclude malicious users under a constrained budget, ensuring the reliability and freshness of the collected sensing data. |
