Design And Analysis Of Hotel Search Recommendation

Along with the development of information technology and Internet, people came tothe era of information overload from the ear of information poor; it’s difficult for users toobtain useful information quickly from vast amounts of information on their own,information utilization decreases as a result. So filtering of information has become theimportant indicator to measure an information system’s quality. Good Systems will filterthe vast amounts of information, present the information users most concern about to users,which will greatly increase the efficiency of the system, and save user the time looking forinformation. Recommender systems came into being in this context, as a supplement totraditional search engines, play an important role in solving the problem of informationoverload.Taking a travel vertical search sites for instance, we expand the research onrecommendation technology on hotel search. After in-depth analysis of a variety ofcommon recommender system, combined with the hotel recommendation typicalapplication, we designed a hotel recommendation system based on hotel similarity. Thisrecommender system will use user’s most recent visits to the hotel to infer his interest, andthen recommends similar hotels. Our system consists of offline module and online module,offline module uses the click logs and hotel information to calculate hotel similarity tables,online module uses user’s recent visits to calculate the recommendation result and collectsuser feedbacks and records system status. In order for the system offline evaluation andresearch, we also designed an evaluation system based on user access time series, anddefine two precisions indices which are hit rate and hit rate accuracy as the mainevaluation indices. The evaluation system treats each user click detail log as accesssequence, with the time window consisting of the recent visits, current access hotel anddestination hotel sliding on the access sequence to simulate and replay the user’s accessand recommendation process with doing associated statistics and calculating theevaluation index. The evaluation system has been used to study the content-based andcollaborative filtering similarity algorithms’ impact on the system, and to explore theeffect of various factors affecting recommendations and ways to improve the systems performance.After the study, we found that similarity algorithms based on collaborative filtering(Amazon’s algorithm and click details conversion rate algorithm) work best, normalizingthe similarity is necessary, and the hotel similarity table should be frequently updated.Using the best training set length, filtering bad data and combined withmulti-recommender engine can effectively improve the system performance. After thecombination use of these improvement methods, as opposed to the original system, the hitrate increased by7%hit rate accuracy is improved by15%.