Algorithms Based On The Weighted Deviation Table Slopeone Improvements

In the recent years, with the rapid development of the internet, the number of e-commerce business websites is becoming larger and larger. There are many music sites, movie sites or book websites which are mushrooming developing. But for each music site, movie site or book website, there are too many categories of goods and commodities information, many users can hardly find useful product information that they want.In recent years, many recommendation algorithms are put forward to make the website recommend goods more accurately and improve the efficiency to lead the users to use the site more efficiently. The Slopeone algorithm is proposed in2005that is based on ratings recommendation algorithm. This algorithm is easy to implemented, having a better scalability, and be able to use a smaller amount of computation to achieve the same purpose as other complex recommendation algorithms.But Slopeone algorithm also has its own limitations, during the recommended score predicting. For example, it cannot handle Data Sparsity very well. When there are too many empty scores, the prediction of scores may have a lower accuracy rate, which may lead to a deviation in results and have influence in final results. This can be improved by pre-processing of the data. The purpose of this paper is to find a way to reduce this deviation. To improve this algorithm, a lot of papers choose the cluster analysis algorithms that tend to have an artificial threshold or a number of centers, which bring new deviations in instead.This paper intends to improve the original SlopeOne algorithm with the use of the similarity between users or the thought of using the weighted method. The similarity of this paper is the similarity between users, which should be reasonable. The weighted method eliminates the effects of artificial factors, which is weighted method’s own advantage. With these two points, our purpose is to get considerable rating prediction results with an acceptable time complexity or obtain better rating prediction results by increasing certain time complexity.