在這網路普及的時代，透過智慧行動裝置(location-based services)查詢附近的店家、地標等POI(Points of Interest)是人們生活中常見的需求。現今人們習慣在Google Map上找尋POI，但並不是所有我們想要的POI在Google Map上都找的到，有些POI只會在網頁上出現但不會出現在地圖上，所以Google Map的POI仍然不足，因此整合多來源的POI，提供一個有效檢索POI的地圖服務系統是本文的目標。人們在地圖服務系統上給予查詢詞，可能是要找多個符合查詢的POI，也有可能是某個特定的POI；在搜尋範圍的考量上，人們可能想找到附近週遭相關的POI，也有可能是不在範圍的POI；POI的排序方式也是個問所在，查詢結果可能需要以距離或相關性來排序。若地圖服務系統的POI相關資訊不充足，那麼效能就會低落。除了透過檢索系統獲得相關的POI，查詢詞擴展也是個可以幫助人們找到相關POI的方法。 在本論文中，我們的系統目標在於提高搜尋結果的精準度，方法可分為兩個部分，第一部分為POI檢索，我們查詢多個搜尋服務的POI結果並設計數個特徵，接著建立預測模組，找出和查詢詞相關的POI。第二部分為標籤推薦，我們將POI描述句彙整成一個語料庫，並經由主題模型分析取得主題詞彙，透過這些詞彙獲得POI的標籤，並建立詞彙和標籤的關係二分圖，最後根據查詢詞做相對應的標籤推薦。 ;With the popularity of internet, the demand of people’s lives is searching local POI (Points of Interest) by intelligent mobile device (location-based services) e.g., stores, landmarks. Today, people used to look for POI on Google Map. However, not all that we want POIs on the Google Map. Some POIs appear on the web page but does not appear on the map, so the POI is still insufficient on the Google Map. Therefore, integrating multiple sources and providing an effective retrieval POI map service is our goal in this paper. People give a query on the map service. They may look for multiple matching query of POIs or a specific POI. In the search scope part, people may want to look for relevant POIs in the neighborhood or out of the scope. How to sort the POI is also a problem. The search results may need to sort by relevance or distance. If the POI Information is insufficient in the map service, the effectiveness would be low. In addition to finding relevant POIs by retrieval system, query expansion is a good method to help people find relevant POIs, In this paper, our goal is to improve the effectiveness of search results. The approach has two parts. The first part is learning to rank. We design several features and get the most relevant POI to query through correlative prediction. The second part is query expansion. After collecting POI descriptions and clustering it by LDA (Latent Dirichlet allocation) , we build a bi-partite graph and give a certain domain of words by graph.
