Research On Mulisource User’s Interest Data Fusion Tree-based Network

With the rapid development of e-commerce, online purchasing has become a mainstream way to shop for existing online store, when they faced variety information, consumers need to spend a lot of time to find their own interest commodity. Under this background, personalized recommendation system came into being, it is considered to be an efficient and consistent with the recommended method commodity to satisfy consumer demand, via solving the problem of e-commerce websites in consumer shopping, it has become the hot research field of information services. It collected user’s behavior, analyzed preference for personal interest, to provide "one on one" personalized service. To ac hieve high-quality personalized recommendation, user interest model need to be done. It is the core of personalized service, so the quality of user interest model is directly related to the quality of the recommendation system.To this, capturing muli-source user interest data fusion is an important way to improve the quality of user interest model. The purpose of this study is aimed at B2 C website environment, traditional collaborative filtering recommendation accuracy is not high question, proposed and implemented based on the user tree network of multi-source user interest data fusion method, to improve the quality and optimize the original method. The full text mainly research contents are as follows:First, by regarding the process of user modeling as research viewpoint, we ran the comparative study on current achievements of user modeling from four aspects, i.e. data collection, data representation, data process and data updating. We classified data collection into two aspects, including data sources and data storage, for obtaining necessary modeling data; classified data representation into two kinds of methods, including semantic representation method and quantitative representation method, for decrypting user interest information; classified data process into two kinds of technologies, including feature weight, clustering, for generating user model by handling user interest information; and classified data updating into three kinds of methods, including time window, forgetting algorithms and hybrid models, for reflecting user interest drift.Second, from the angle of user’s shopping process, it summarized four kinds of interesting attribute factor that can reflect consumer interest in maximum extent, they are: hits, collection, shopping cart, order goods. We gave to specific quantization of each index factor calculation, and set up corresponding rules for static users interested in weight. Considering the user interest changes, design variety interest value to make up for the inadequacy of the static system. For each user’s purchase behavior, the user’s interest is divided into long-term interests and short-term interests, and offer different interesting index attenuation ratio. Using different attenuation method can better improve recommendation accuracy, besides improve the system reliability.Finally, based on cat store large amounts of real user’s access data, the experiment trained with each user’s interest model, and calculated each user corresponding to the long-term interests and short-term interests. This paper had three contrast experiment, The value of the first group to explore each attribute index factor; A second group of attenuation for cycle model with exponential decay model does not distinguish between interest cycle forecast accuracy experiments; A third group of classic collaborative filtering algorithm and the contrast experiment of attenuation filtering algorithm proposed in this paper. Experimental results show that the recommendation of multi-source data fusion user interest effect is superior to the classic collaborative filtering recommendation effect. One group is to explore value of each attribute index factor. A second group is that cycle attenuation experiment method compared with the same all of the user’s interest in using the same attenuation experiment method by forecasting accuracy. A third group is that the classical collaborative filtering algorithm compared with the change collaborative filtering algorithm. The experimental results show that the recommendation of multisource data fusion user interest effect is superior to the classic collaborative filtering recommendation effect.