CN102693225A - Internet consultation collaborative filtering method - Google Patents

Abstract

An internet consultation collaborative filtering method is provided, comprising: a user retrieval requirement is subjected to an initialization process; and a retrieval result obtained by a collaborative filtering algorithm is output, wherein the collaborative filtering algorithm can be one selected from a Clustering method and a grouping algorithm.

Description

Internet consult is coordinated filter method

Technical field

The present invention relates to information retrieval field, relate in particular to a kind of internet consult and coordinate filter method.

Background technology

The world today, along with development of internet technology, such as WWW, Netnews, various information sources such as Mailing list continue to bring out, make people might be from more information source acquisition of information.Meanwhile, a large amount of information has caused the blast of information, makes people have to spend a large amount of valuable time in order to obtain required information, thereby information acquisition being become be close to has lost meaning.

For addressing this problem, various way and scheme have appearred.But also exist simultaneously common problem: though the information that goes for do not exist because the user can't judge and also retrieve perversely; The information of wanting to obtain exists really, yet can not find these information owing to method is improper; In the information retrieval process, the unnecessary information of collecting as a large amount of floods causes Useful Information to be submerged; Up to now, obtaining information from the internet has information retrieval, information several method such as to filter and browse.

And the common way of independently carrying out information acquisition, even caused the user who is concerned about same content can't realize that also information has/shares.And in real world, for the common topic of care, the user of common content, a sixty-four dollar question carries out information exchange efficiently exactly and shares.But the realization of very regrettably traditional gimmick is this to be had/share is impossible.Independently information acquisition is the mortal wound of existing various main retrieval softwares each other, is a kind of worthless mode.Internet development presses for a retrieval support system that is used for information gathering, has intelligence.

For solving the above problems, a kind of so brand-new method has appearred coordinating to filter.Coordinate the knowledge that filtration method has made full use of other people and carry out information acquisition.And these knowledge have usually comprised indirectly even the important information of directly dealing with problems.This way has significantly reduced required time and the workload of collect intelligence.More particularly, according to user's hobby and requirement, system provides/recommends various information automatically, also is the method for recommendation service (Recommence Service).

Coordinating filter method and be based on information circulation propagation, is the method for the information acquisition of target to improve collection efficiency.The information circulation is meant

(1) seeks the information of dealing with problems automatically for the user who holds different problems;

(2) to the user who holds same problem recommendation service is provided;

(3) need to seek the customer group of paying close attention to certain particular problem, and then obtain being the information necessary of dealing with problems.2. then be that these users' characteristic is concluded, accomplish the exchange of information.Coordinating filter method then combines above-mentioned process together naturally.So we can do following definition for coordinating to filter.

The definition of coordinating filtration (Collaborative Filtering) is: from the mankind's information acquisition activity; Conclusion takes out its pairing hobby; Be concerned about, forms such as intention consciousness, and realize through the form that the collect intelligence that obtains and conclusion take out; The mankind are classified, realize the means of the information exchange between the similar mankind.

As coordinating a kind of of filter method, for realizing from the information as the flood, to extract the necessary information of user, the standing use of commending system (RecommenceSystem).In addition, be the automatic tracking and the judgement that can realize that the person of utilization likes, the research and utilization of Agent system (Agent system) technology and artificial intelligence technology also has very big potentiality.

Following table has been listed the principal feature of coordinating filtration method.

Can know by last table, look on the bright side of things and send out successful goods, just need maximize favourable factors and minimize unfavourable ones, on aforesaid technological inscape, implement careful adjustment.Can say so, how various technology essential factors organically are combined into a complete system, be native system commercialization key of success.

Summary of the invention

Be problem and the defective that solves above-mentioned existence, the invention provides a kind of internet consult and coordinate filter method that be applicable to the identification of any attribute in arbitrary data storehouse, its search method comprises:

According to the user search demand, it is carried out initialization process;

According to coordinating the result for retrieval that filter algorithm output obtains according to algorithm.

The beneficial effect of technical scheme provided by the invention is: through the analysis to word or statement in the user search demand; The attribute information of predictive user query terms or statement; And inquire by classification according to word or the different attribute information of statement; Return more accurate, the result for retrieval that user satisfaction is higher.

The algorithm of coordinating to filter can be divided into 3 types.

One of which, Active Collaborative Filtering.Utilize this technology, can specify each other between the user who understands mutually, (have certainly safe and secret on restriction) obtains Useful Information (comprising Email) each other.Can keep synchronous with the expert within this field.

Its two, Automated Collaborative Filtering and Feature Guided Automated Collaborative Filtering.In many ways the result who analyzes from the angle of efficient and precision sees that Feature Guided Automated Collaborative Filtering holds a safe lead.

Its three, Content-Based Collaborative Filtering.At first be to be object, and then consider other media with article information.

Among all algorithms of Cluster, the precision of Wood method and group average method is better.

Dividing the set of calculated aspect, is to utilize similar algorithm basically.Jaccard ' the s Coefficient method in the employed similar algorithm of dividing into groups, average least square Furthest Neighbor, the ratio of precision of improvement two-value Furthest Neighbor is more satisfactory.

The purpose of Clustering method and grouping algorithm is identical.The characteristics of Clustering are that precision is high, but processing speed is slower.Grouping algorithm is then on the contrary.Can come the use of these two kinds of methods of balance by system control parameters, adjust flexibly according to the scale of system.

Differentiating similar object has 4 kinds, i.e. user's (attribute) similar to user's (attribute), and article (attribute) is similar to article (attribute), and user's (attribute) is similar to the similar and key word of article (attribute).The object of judging is different, and evaluation result also has nothing in common with each other.Similar between the class Sihe article between the user press Jaccard ' s Coefficient method, and average least square Furthest Neighbor is improved two-value Furthest Neighbor Furthest Neighbor, the order of Pearson's correlation method, and computational accuracy is more satisfactory.And concerning user and article similar, then by improvement two-value Furthest Neighbor, Jaccard ' s Coefficient method, average least square Furthest Neighbor, the order of Pearson's correlation method, for computational accuracy in order.The similar consideration of key word goes to realize with statistical method.In addition, owing to must confirm to provide the thresholding of scope, so the same distance algorithm is compared, related algorithm will be fit to manyly.If, believe also and can further improve the recommendation precision with the use that combines of top several method.

To the general user following service is provided mainly:

Recommendation service: searching and object user's similar users, recommend the article that the object user did not visit to it;

Similar article service is provided: the guide look with the similar article of object article is provided;

The filtering services of result for retrieval: on the basis of result for retrieval,, the result is selected according to the user's who implements retrieval hobby.

New recommendation service: according to each user's demands of different, in certain period, newly offering the user to information to information;

User characteristics specified services: allow system registration or the deletions such as word/article of user with its concern;

Similar key word service is provided: system can provide and retrieve with the similar key word complete list of key word;

Popular webpage service is provided: the address that welcome webpage is provided;

The individual character advertisement service is provided: concerning Internet service merchant (ISP), can be according to user's characteristics, hobby and just right advertisement initiatively is provided;

Be the convenience of bookkeeping, also provide function abundant service simultaneously to the system manager.

About architecture:

Employing is suitable for the Java language of internet WWW service system exploitation, with the form exploitation of Servlet;

Coordinate filtering system and belong to a kind of of Multi-Agent system, so must have the characteristic of its dispersion treatment.Standard-CORBA (Common Object Request Broker Architecture) that native system will adopt the computing machine dispersion technology realizes decentralized processing;

Adopt DBMS that information is managed.And utilize other part of JDBC and native system to link to each other;

The dissection process of natural language can take the considerable time of system, will adopt C Plus Plus as far as possible, provides with the form of built-in function (LIB).Prepare to use JNI with being connected of other part of native system;

Native system allows to set in advance user's characteristic information, even simultaneously as not doing any setting, can be according to the track of user's operational processes, and system will infer hobby and the migration thereof that the user automatically.But the migration of not preparing to influence with simple accessing operation or search key consumer taste simultaneously, plan provides control information and control corresponding to handle for this reason.On the other hand, also with the scale and the load of taking into account system;

The supvr can carry out simultaneously article in enormous quantities collection, preserve to handle, extract the attribute of article and to its management.Also can from result for retrieval, realize above-mentioned processing through general user's accessing operation.But whether preserve, can control through access times.

The above is merely preferred embodiment of the present invention, and is in order to restriction the present invention, not all within spirit of the present invention and principle, any modification of being done, is equal to replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (2)

1. an internet consult is coordinated filter method, comprising:

According to the user search demand, it is carried out initialization process;

According to coordinating the result for retrieval that filter algorithm output obtains according to algorithm.

2. internet consult as claimed in claim 1 is coordinated filter method, wherein coordinates filter algorithm and can be a kind of in Clustering method or the grouping algorithm.