Research On Accurate Recommendation Of Medical Information Based On User Portrait

The traditional interrogation patterns include waiting for registration, waiting to see a doctor, taking inspection report, invoicing and taking medicine etc. It's a length and jumbled process especially for patients. Sometimes the best time to treat the disease disappear with waiting. To solve the problem, diagnosing illness through internet develop rapidly, which makes the medical procedure much more convenient and effective. However, the traditional keyword-based searching couldn't meet the needs of patients well due to data accumulating in the age of big data. As a result, how to deal with the huge amounts of information through data mining effectively has become a common concern.In this thesis, data portrait theory and recommendation system related technologies are used to find a recommendation algorithm of the characteristics of high efficiency, high stability, strong real time and personalization. This algorithm can implement a precision recommendation system of doctors and illness information for the Internet users and provide them with an automatic or semi-automatic triage system, so as to achieve accurate transfer of medical information. The studies were summarized as follows:(1) A medical portrait library was set up based on data portrait theory. This part of work involved designing medical portrait model, generating algorithm, updating algorithm, analyzing and applying patient information, and finally setting up a medical portrait library.(2) A set of recommendation algorithm based on the medical portrait library, SVD collaborative filtering and the Tag attributions of the portrait was formed. In this part of work, SVD(Singular Value Decomposition) based collaborative filtering was introduced to reduce the problem of data sparseness. Meanwhile, the research of medical portrait was applied to set up the recommendation algorithm. The key step to realize algorithm is the similarity's measure in collaborative filtering. In this paper, two kinds of similarity's measure algorithm were designed: the traditional similarity's calculation and similarity's calculation that based on label attribute preference. A dynamic keyword w was adjusted repeatedly to balance this two methods. As a result, an appropriate w was chosen to efficiency of performance improvement and the precision of the recommend system.(3) Finally, a medical recommendation system based on B/S model was realized. A better communication between patients and doctors could be realized through the browser interfaces showing interrogation window, presentation of the portrait and recommended results. The individual portrait was based on the user who logged in the system. The recommended results was provided according to the portrait. Thus a real personalized recommendation system was achieved.