Research on Web page weighting method based on keywords

Intermediary transaction http://www.aliyun.com/zixun/aggregation/6858.html ">seo diagnose Taobao guest cloud host technology Hall

After clustering, the user transaction pattern obtained by the keyword sequence forms the different pattern features of expressing the user's personalized information demand. Compared with the keyword sequence, the user transaction pattern after clustering has obviously fewer quantity and more explicit personalization characteristics, so the personalized information recommendation activity can be carried out with this user transaction clustering mode.

At present, there are two main methods of personalized information recommendation in search engine: One is query improvement, the other is personalization page weight. The former mainly use to change the user's actual search keyword content to express the user's personalized information needs, and the latter is mainly based on the weight of the page to the results of the order to express the user's personalized information needs. The recommendation method based on personalization page weight and the recommendation method based on query improvement have many advantages, mainly in the following aspects:

First, through practice, such as PageRank, such as Web page weight value is a more effective reflection of the objective importance of the index, and the corresponding algorithm has a technical ease.

The second is that the algorithm mainly solves the problem of the objective importance of Web pages, and can play a role in the ranking of Web pages ' relevance. In other words, the algorithm can put the Web pages that meet the personalized requirements of users to the end of the collection of results pages. In fact, this helps users get the personalized information they need.

Finally, the relevant web page weight calculation work need not be online, just use the off-line phase of the storage data can be calculated, so that the user can effectively save the time required to query the cost.

The idea of recommendation algorithm based on the weight of personalized Web pages is based on the traditional Web page weight algorithm, by modifying and adding the specific parameters in order to express the characteristics of different users ' personalized requirements, so as to calculate the weight value of the Web pages peculiar to different users, and when the user queries, Use this value to calculate the relevance and precedence of the Web page.

The more common way of personalized Web page weight is personalized PageRank method. The traditional PageRank is a sort of relevance ranking technique used in query results page, it calculates the weight value of different pages by the link person and the link of the Web page, and realizes the Web page sort. There are many kinds of derivative types in this algorithm, and the main aim is to make further information expression to the result. Among them, the most common approach is to use the personalized PageRank vector to express the personalized information needs of different users, and use this vector to calculate the relevance of the Web page, resulting in personalized search results for specific users.

Personalized PageRank algorithm mainly based on personalized PageRank vector to make the results of the Web page to create a specific user preferences. Among them, many algorithms are based on the web-based graph theory algorithm, the most common models have Markov model and so on. For Markov models, many different specific types, such as Order Markov chain model, Gaushem Markov chain model and hybrid Markov chain model, have been proposed. The first order Markov chain model can give a simple descriptive method to the sequence dependence, but it does not consider the long term memory characteristic of the network surfing behavior; the Gaushem Markov chain model can predict the navigation path more accurately, but it also has the tradeoff of coverage and computational complexity as the dimension increases, Moreover, the complex model usually requires a considerable training set, and the hybrid Markov chain model combines the Markov models of every order and needs more resources in preprocessing and training.

Obviously, the algorithm here is relevant to the chosen model, depending on the tradeoff between simplicity and correctness, the selected model type, or even some other models, are mostly based on tree-like navigation graph data mining algorithms, such as clustering, sequential model mining, frequent item mining, and so on.

Although there are a lot of personalized PageRank methods, it is mainly divided into two kinds: one is to modify the weight value of the Web page directly based on the hyperlink relationship, the other is to add the correction parameter to the traditional PageRank formula to reflect the user's personalized request.

This article by the www.q322.com webmaster feeds!