以標題為基礎的產品搜尋：中文電子商務平台為例

Title-based Product Search - Exemplified in a Chinese E-commerce Portal

陳健文(Chien-Wen Chen)

指導教授 ： 鄭卜壬

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2010年)

https://doi.org/10.6342/NTU.2010.01131

摘要

近幾年來，線上購物越來越流行。一個線上購物網站，例如eBay，在任何時候同時會銷售超過7500萬種產品。我們必須要幫助買家有效率地找到想買的產品。一般以關鍵字為基礎的資訊檢索系統似乎可以幫助使用者來搜尋產品。但很不幸的，我們從真實的產品搜尋查詢字紀錄中觀察到，查詢字通常都很短，因此較難去了解買家的意圖。更糟的是，一個描述產品的文件中，常常會出現和相關的產品有關的文字。一般的資訊檢索系統很難去分辨出具有代表性的文字。
因此，我們提出一個想法來了解產品文件標題中每一個字在語意上所扮演的角色。我們用一個監督式的機器學習方法，來預測每個字的語意類別。使用這個預測模型，我們修改傳統的語言模型以增加搜尋結果的相關程度。在我們的實驗中，我們發現了一些在中文平台的賣家的行為習慣，並且將我們的搜尋結果和傳統的搜尋方法產生的搜尋結果做比較。相比較的方法包含有向量空間模型和語言模型。從真實的產品文件和查詢字中，我們的方法在搜尋結果的準確度上有顯著的進步。

關鍵字

資訊檢索 ； 自然語言處理 ； 產品搜尋 ； 電子商務

並列摘要

On-line shopping has become more popular in recent years. There are too many products in an on-line shopping website. Take eBay for example, their platform sells more than 75 million kinds of products at any time. We need to help buyers to find products they want in an efficient way. A keyword-based information retrieval system seems suitable for searching products. Unfortunately, we observe from real world query logs and find that queries for product search are usually very sho rt. Therefore, it is hard to realize buyer's intention. What is worse, a document described a product may have lots of words of related products. It is hard for an information retrieval system to distinguish representative terms from other noisy terms.
Hence, we propose an idea to realize semantic types of each term in product document titles. A supervised learning method is used to predict semantic type for each term. Using the prediction model, we modify Language Model to improve the relevance of search results. In our experiments, we find some interesting behaviors for Chinese sellers and compare the ranking result to the results of traditional methods including Vector Space Model and Language Mod el. Our methods have significant improvement in precision in real world document collection and query collections.

並列關鍵字

information retrieval ； natural language processing ； language specific IR ； domain-specific IR ； product search ； e-commerce

參考文獻

Available at http://th.nielsen.com/site/documents/

[2] C.-C. Chang and C.-J. Lin. LIBSVM: a library for support vector machines, 2001.

Software available at http://www.csie.ntu.edu.tw/˜cjlin/libsvm.

choice: Query expansion based on the layer-seeds method and its application in ecommerce.

DC, USA, 2004. IEEE Computer Society.