微網誌上行銷智慧流行議題監控之研究

Abstract

隨著使用人口的快速增加,微網誌已經成為大量收集消費者意見很好的來源。然 而,微網誌有許多不同於傳統網路資訊來源的特性。因此,面對微網誌不同的特性以 及更為大量的資料,本研究提出一個數值化整合微網誌意見的系統框架,以期能在微 網誌平台上發掘出良好的行銷相關的資訊與知識。我們提出的框架能夠處理以下四個 問題:1)主題偵測 2) 意見情緒分類 3) 使用者可信度評估 以及 4)數值化整合。我們分 別以支援向量機(Support Vector Machine, SVM)與部分詞格式(Meronym Patterns) 處理意見情緒分類與主題偵測並以社會網路分析的指標來衡量發言者可信度。最後上 述資訊將進行加權整合,提供對於各主體的不同主題整合的意見分數。我們在目前最 大的微網誌平台twitter上進行實驗,驗證此系統框架的有效性。我們發現,進行意見整 合時,考慮使用者可信度以及訊息品質,能夠提供更良好的整合結果。

With rapid growing popularity, microblogs have become great sources of consumer opinions. Confronting unique properties and massive volume of posts on microblogs, this thesis proposes a summarization framework that provides compact numeric summarization for microblogs opinions. The proposed framework is designed to cope with four tasks: 1) trendy topics detection, 2) sentiment classification, 3) credibility assessment and 4) score aggregation. Support Vector Machine and relation patterns are used for sentiment classification and topic detection respectively while microblog specific indicators are adopted for user credibility assessment. A weighted aggregation is then applied on above information to calculate scores toward relevant topics of a user query. An experiment is held on Twitter, the largest microblog website, for proving the efficiency and correctness of the proposed framework. We found the consideration of user credibility and opinion quality is essential for aggregating microblog opinions.