The course of mode of user general affairs that uses keyword alignment to get gets together kind later, formed the different pattern feature that conveys requirement of user individuation information. Compare with keyword alignment photograph, get together mode has kind of user work after apparently fewer amount and more specific individuation feature, use this user general affairs to get together so kind of mode is OK undertake next personalized information recommends an activity.

The common personalized information in searching engine at present recommends means to basically have two kinds: One kind is inquiry improves; one kind is personalized webpage weight. Former advocate the personalized information requirement that if use a change,the keyword content that the user searchs actually conveys an user, and latter basically is the personalized information demand that undertakes to result webpage according to webpage weight sort will convey an user. The commendation that is based on personalized webpage weight the method recommends methodological photograph to compare with what be based on inquiry to improve, have a lot of advantages, main show is in the following respects:

It is course practice proof, the webpage weight value such as such as PageRank is webpage of a kind of relatively effective report the index of objective importance, at the same time corresponding algorithm has sex of technical easy travel.

2 it is algorithm basically solved a webpage the evaluation problem of objective importance, can be related the webpage spend action of the play on sort. That is to say, the webpage that this algorithm can ask contented user individuation is put in result webpage aggregate foremost end. In fact, this more the personalized information that conduces to an user getting needing.

Finally, relevant webpage weight computation works need not online undertake, need to use the memory data of phase leaving a line to be able to be calculated only, can save an user effectively to inquire the time of a need pays expenses thereby.

The commendation that is based on personalized webpage weight is algorithmic the thought is the webpage weight algorithm in the tradition over the foundation, through be revised reasonably and adding the specific parameter that uses among them, in order to convey the personalized demand feature of different user, calculate thereby the webpage weight value with a different user peculiar place, it is when user inquiry, use this value to calculate of the webpage relevant spend and first step.

Pattern of relatively familiar personalized webpage weight is personalized PageRank method. Traditional PageRank is a kind those who be used at inquiring result webpage is relevant spend sort technology, the catenary person that it asks through the webpage and catenary give a relation to calculate the weight value of different webpage, achieve webpage sort accordingly. This kind of algorithm already had a variety of deriving at present type, main purpose is the information expression with farther to be being done as a result with a view to. Among them, most common practice is the personalized information requirement that vector of use individuation PageRank will come to convey different user, what use webpage of this vector computation is relevant degree, produce the individuation that is aimed at specific user to search a result thereby.

Personalized PageRank algorithm basically makes according to personalized PageRank vector result webpage produces the preference character of pair of specific users. Among them, a lot of algorithm are a foundation in order to be based on the graph of Web by algorithm, most common model has Markov model to wait. To Markov model, people had offerred a lot of different specific types, wait like model of model of model of first-order Markov catenary, high-ranking Markov catenary and mixture Markov catenary. Among them, although model of first-order Markov catenary can be depended on to alignment give out a simple descriptive method, but its lasting effect without behavior of consideration network surf remembers model of catenary of diagnostic; high-ranking Markov to be able to forecast navigation method well and truly, but it also can spend grow in quantity as dimension and generation covers degree of balance issue with computational complexity, and this kind of complex model asks quite big training part; mixed Markov catenary model to combine each normally rank Markov model, mix in pretreatment more resource also need when training.

Apparent, the algorithm of here and chosen model are relevant, want a basis to decide what to select model kind to the balance circumstance of simple and easy sex and validity, it is a few other models even, these other model most use the data that is based on texture of picture of the navigation that establish record to dig algorithm, if get together kind, series model mining, frequent a mining.

Although the personalized PageRank means that people offers now has a lot of, but main component is two kinds big: One kind is to revise the webpage weight that is based on relation exceeding cable length to get to be worth; directly another kind is the personalized requirement that adds correction parameter to mirror an user on traditional PageRank formula. (Think of.cn of science and technology of network of 100 million Europe)