CN1645395A - Method for discovering user interest in e-mail flow and transmitting document effectively - Google Patents

Abstract

The invention relates to the field of computer technology, in particular to a method for discovering user interests in email streams and effectively pushing documents accordingly. Members of the same scientific research team often repeatedly search and download the same documents due to their overlapping research fields. When the issues concerned by members change, their interests will be updated in time with the e-mails they send and receive, and the correct documents can always be pushed to members according to their interests; perform semantic analysis on the documents in the team document database, and push them to members according to the analysis results Documents that are consistent with their interests semantically ensure that the documents pushed to the user are exactly what the user needs, accurate and effective; members only need to upload the document to the team's document database, and the program can complete the analysis and push of the document, realizing The sharing of the document among members is simple and easy.

Description

A method for discovering user interests in email streams and effectively pushing documents accordingly

technical field

The invention relates to the field of computer technology, in particular to a method for discovering user interests in email streams and effectively pushing documents accordingly, for semantic understanding, text classification, document sharing and email streams.

technical background

The research fields of different members of a scientific research team usually overlap. On the one hand, they often repeat the search and download operations in order to obtain the same document, resulting in a waste of manpower and financial resources; on the other hand, they often exchange information by email, and sometimes Also send valuable documents as attachments to other members, which can achieve document sharing among members to a certain extent, but there are still the following problems:

First of all, there is no guarantee that each member is willing to send other members the documents they need, so it is impossible to fundamentally avoid repeated operations by team members to obtain the same documents.

Second, even if each member is willing to send other members the documents they need, there will still be situations where a member's interests often change over time, and other members may continue to contribute without knowing this change. He sends documents that he no longer needs now, but he does not send him newly needed documents; it is difficult for a member to accurately grasp the interests of all other members, so that the document cannot be pushed to all members who need it, and the document cannot be realized. fully shared.

In order to realize the full sharing of scientific and technological documents in the team, the present invention first extracts the interests of each team member in scientific research work, and then regularly pushes relevant documents to the team members according to the interests of the members. Accurately extracting the interests of team members is the basis for fully realizing the sharing of technical documents among team members. In the process of sending and receiving e-mails, an e-mail flow is formed among team members, and the concerns of each member can often be reflected through the e-mails sent and received, so the interests of team members can be extracted from the e-mail flow . Based on the existing email function, the invention extracts user interests from email streams among team members, ensuring the premise that documents are fully shared among team members. The basic idea is that where the e-mails sent and received by the members are concentrated is the place where the members' research work is concentrated: first, the e-mails between members are saved in the database, which process eliminates the interference of spam; second, using natural language learning The method to obtain effective emails that can provide useful information for describing user interests; then, divide the research field related to the team into smaller subfields, and classify the effective emails on this basis; finally, according to the effective emails The distribution in each subfield, using the set of subfields concerned by members to represent user interest. Considering that user interests may change after a long period of time, the time factor is introduced into the interest extraction process. User interests will be updated in time with the generation of new emails and the passage of time. Push documents according to user interests to ensure that documents can always be pushed To all team members who need the document, it will neither be sent by mistake nor missed.

The present invention uses the set of interest points describing the semantics of the subfields as a template to divide documents into subfields with similar semantics. Documentation is semantically what the user needs, accurate, and valid.

If a team member wants to share a document with other members, they only need to upload the document to the team's document database to realize the understanding and push of the document. Most team members can accept the simple upload operation, which is largely realized Document sharing among team members avoids their cumbersome duplication of operations.

Contents of the invention

The purpose of the present invention is to provide a method for discovering user interests in email streams and effectively pushing documents accordingly, thereby effectively utilizing team resources and fully realizing the sharing of scientific and technological documents among team members. The steps of this method are as follows: firstly, store emails between team members into the database; then, extract user interests from the email flow among team members, and when the issues that members pay attention to change, their interests will be sent and received with them Emails are updated in time, and the correct documents can always be pushed to members according to their interests; and semantic analysis is performed on the documents in the team document database; finally, on the basis of document semantic analysis, documents consistent with user interests are pushed to team member.

This method mainly includes the following points: forward the e-mail among team members to a certain fixed account through the function provided by the e-mail server program, regularly execute the mail collection program, and this program decodes the e-mail in the fixed account and decodes the result Save in the email database to complete the automatic storage of emails. Most spam emails come from unfamiliar email addresses. This invention only saves emails between members, which eliminates the interference of spam emails when extracting user interests ; Only consider the interests of team members in scientific research, use the method of natural language learning to divide emails among members into valid emails and invalid emails, and obtain valid emails that can provide useful information for describing user interests. Based on the extraction of user interests, the accuracy of user interests is ensured; each research field related to the team is subdivided into sub-fields, and the background knowledge and semantics of the sub-fields are expressed through the prior knowledge sets and interest point sets of the sub-fields; The similarity calculation between effective emails and prior knowledge sets realizes the classification of effective emails. The subfields in the user’s effective email set are exactly the subfields in its research work. Therefore, according to the distribution of user’s effective emails in each subfield Extract user interests, which are expressed as a collection of sub-fields they care about; user interests may change over time, and the ability to describe user interests in emails should also decrease with the growth of their existence time, introducing time into The process of extracting user interests, when the user's work focus shifts, their interests are also adjusted in time, so the document can always be pushed to all team members who need the document, neither wrong nor missing, ensuring the team The premise of fully sharing scientific and technological documents among members; using the set of interest points describing the semantics of the sub-fields as a template, the documents are divided into different sub-fields according to the semantic similarity between the documents and each sub-field, and based on this, the documents are pushed to The set of concerned subfields includes users in the subfield to which the document belongs, which semantically ensures that the pushed documents are what the user needs, accurate and effective. Team members only need to upload the document to the team's document database to understand and push the document. Most team members can accept the simple upload operation, making document sharing among team members simple and easy.

Technical solutions

The present invention is a method of discovering user interests in email streams and effectively pushing documents accordingly. In this method, firstly, each research field related to the team is subdivided into subfields, and a priori knowledge set representing the background knowledge of the subfield and a set of interest points describing the semantics of the subfield are constructed; Store emails in the email database, and extract effective emails that can provide useful information to describe user interests, and team members can also upload valuable technical documents to the document database. Then, the effective emails are divided into the subfields with the highest similarity between the prior knowledge set and the user interests are extracted according to the distribution of valid emails in each subfield, and the interest point set in the subfields is used as a template to analyze the information in the document database. Documents are semantically analyzed and categorized. Finally, the document pushing program pushes the documents consistent with the user's interests to the team members according to the user's interests and the results of document classification.

This program mainly includes the following technical indicators:

1. Email auto-repository between team members

First, build an e-mail database, each record of the database stores an e-mail, and automatically forwards e-mails between team members to a fixed account through the e-mail server program; then, run the e-mail collection program regularly, the program Decode the emails in the fixed account, and store the decoding results in the email database to realize the automatic storage of emails. Spam usually comes from unfamiliar e-mail addresses, because only e-mails between members are saved, and the automatic storage process of e-mail itself realizes the filtering of spam.

2. Extract valid email

The present invention only cares about the user's interest in scientific research work, so only the e-mails related to the scientific research work content are effective, and the effective e-mails that can provide useful information for describing the user's interests are extracted from the e-mail database through the learning method of natural language .

3. Refine the division of scientific research fields, and establish prior knowledge sets and interest point sets in subfields

Subdivide the research field of the team to obtain a set of subfields related to the team. Establish prior knowledge sets and interest point sets for each sub-field, respectively representing the background knowledge and semantics of the sub-field. The elements of the prior knowledge set consist of two parts: keywords representing the main content of the subfield and the impact factors (description ability) of the keywords on the subfield. The interest point set is composed of semantic link network corresponding to the interest points contained in the sub-field, and a semantic link network describes the semantic information of an interest point.

Establish the prior knowledge set of the sub-field to represent its background knowledge, and classify the effective emails by calculating the similarity between the effective emails and the prior knowledge sets of each sub-field. According to the distribution of effective emails in each sub-field, use member A collection of concerned subfields to represent user interests.

Construct a set of interest points describing the semantics of the subfield, use this as a template to divide the document into subfields with similar semantics, and the document pusher will push the document to the members who are concerned about the subfield to which the document belongs, semantically guaranteeing the push to The user's document is exactly what the user needs. Team members only need to upload the document to the team's document database, and the program can complete the push of the document, which is simple and easy.

4. Obtain user interests according to the classification results of valid emails

Determine the subfield to which each email belongs through the matching calculation of valid emails and subfield prior knowledge sets, and realize the classification of valid emails; based on the classification results of valid emails, according to the effective emails related to members The distribution determines the set of sub-domains that members are currently interested in, by which user interest is represented. The basic idea is that a subfield in a user's e-mail set is also a subfield in his research effort set.

5. Timely update user interests

User interests tend to change over time, so the ability of e-mail to describe the user's current interest should decrease with the increase of its existence time. This method introduces the time factor into the process of extracting user interest. When the user's concern changes At the same time, their interests will also be adjusted, and documents will be pushed according to user interests to ensure that documents can always be pushed to all team members who need the documents, neither by mistake nor by omission.

Considering that user interests may change after a long period of time, the time factor is introduced into the interest extraction process, user interests will be updated in time with the generation of new emails and the passage of time, and documents are pushed to users according to user interests to ensure that documents can always be retrieved Push to all team members who need the document.

6. Judging the subfield of the document based on semantic analysis

Using the interest point set of the sub-field as a template, the semantic analysis of the documents in the document database is carried out, and the documents are divided into sub-fields with similar semantics, which guarantees the accuracy of document classification semantically. Regularly perform semantic analysis and segmentation of documents newly added to the document database.

7. Push documents according to user interests and document classification results

Regularly run the document push program, which pushes the documents in the document database that match the user's interests to the corresponding team members by email according to the user's current interests. Push documents according to user interests to ensure that the correct documents are always pushed to team members; push the results of document semantic analysis instead of simple keyword matching results to users to ensure that the pushed documents are semantically what users need Yes, accurate and effective.

Description of drawings

FIG. 1 is a flow chart of the method for discovering user interests in email streams and effectively pushing documents accordingly according to the present invention.

Fig. 2 is a representation diagram of a semantic link network and its adjacency matrix of the present invention.

Fig. 3 is a flowchart of document understanding of the present invention.

Detailed ways

The present invention is a method of discovering user interests in email streams and effectively pushing documents accordingly. This method subdivides the research fields related to the team into smaller sub-fields, and establishes a priori knowledge set and interest point set for each sub-field to represent the background knowledge and semantics of the sub-field, and the user's interest is the sub-field of interest collection. First, save emails among team members to an email database, and extract valid emails that involve scientific research information. Then, divide the valid emails into the subfields with the highest similarity between the prior knowledge set and the classification of valid emails; according to the classification results, calculate the distribution ratio of the valid emails sent and received by each member in each subfield , add the subfields whose distribution ratio is greater than the threshold to the set of subfields that the user cares about, and obtain the user's interest. At the same time, the documents in the team document database are divided into subfields with similar semantics through the semantic analysis of the documents by using the interest point set of the subfield as a template. Finally, the document push program pushes relevant documents according to the user's interest. The specific implementation method is to push the document in the document database to users whose concerned subfield set includes the subfield to which the document belongs in the form of email attachments.

Fig. 1 is the implementation flowchart of the present invention, mainly comprises following four parts:

1. Automatically store e-mails and extract valid e-mails

1. Build an email database

Team members use a unified email server and server program (such as: WebEasyMail), and create a database file (such as: mail.mdb, Hereinafter referred to as e-mail database) to save e-mail information between team members. Each email is stored as a record in the email database, which contains six fields. The names and meanings of each field are as follows:

Sender: Email address of the sender

Recipient: The email address of the recipient

Cc: email addresses to copy

Sent Time: The time the email was sent

Subject: The subject of the email

Body: The body content of the email, for those longer than 255 characters, stored in the form of object concatenation and embedding

2. Automatic email storage

First, through the service provided by WebEasyMail, all emails between team members are automatically forwarded to a fixed account (such as: an account whose user name is group). The emails of this account are saved in a fixed directory of the mail server (such as: C:\WebEasyMail\mail\group, hereinafter referred to as the undecoded email directory). Spam in the traditional sense usually comes from email addresses that users are not familiar with. This process only collects emails between team members, which eliminates the interference of spam on the user interest extraction process.

Then, run the mail collection program (such as: MailGatherer) written regularly (such as once a day) to realize the automatic storage of e-mails. The program reads each e-mail in the undecoded e-mail directory in turn, analyzes the e-mail header, decodes the e-mail body, and saves the decoded e-mail information into the corresponding field of the e-mail database file; moves the processed e-mail to In another directory of the email server (such as: C:\WebEasyMail\mail\group_deleted, hereinafter referred to as the decoded mail directory), it will not be processed the next time MailGatherer is run. Run MailGatherer regularly.

3. Extract valid email

Although spam in the traditional sense has been filtered out in the previous step, not all emails stored in the email database can provide effective information for describing user interests. We refer to emails that reflect user interests as valid emails, and emails that do not reflect user interests as invalid emails. Emails related to the research content of the team are valid emails; while jokes or event notifications often sent among team members are invalid emails. Only members’ interests in scientific research work are considered here. In order to obtain accurate user interests, valid emails must be extracted from the email database, which is realized through natural language learning.

表示： First, select a certain number of valid emails and invalid emails as the training set C ₁ of valid emails and the training set C ₂ of invalid emails respectively, and obtain the standard vectors of valid emails and invalid emails by the following formula and

express:

${\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; c</span>}}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 16</span>}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; |</span>}\underset{\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\frac{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; e</span>}}}{<span\; class="notranslate">\; |</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 4</span>}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span>}\underset{\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\frac{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; e</span>}}}{<span\; class="notranslate">\; |</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

${\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; c</span>}}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 16</span>}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; |</span>}\underset{\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\frac{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; e</span>}}}{<span\; class="notranslate">\; |</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 4</span>}\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{<span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; |</span>}\underset{\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; C</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}\frac{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; e</span>}}}{<span\; class="notranslate">\; |</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

是 的向量长度；|C₁|和|C₂|分别是C₁和C₂的训练样本数，即，包含的电子邮件数。然后，计算电子邮件数据库中电子邮件e的向量表示 与标准向量

和 的相似度，计算方法如下：in, $\stackrel{\&Right\; Arrow;}{e}=(e_1,e_2,...,e_{\left|f\right|})$ is the vector representation of e-mail e, and e _i is the number of occurrences of keyword w _i in the subject and body of e-mail e;

yes |C ₁ | and |C ₂ | are the number of training samples for C ₁ and C ₂ , respectively, that is, the number of emails included. Then, compute the vector representation of email e in the email database with the standard vector

and The similarity is calculated as follows:

$\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; e</span>}}<span\; class="notranslate">\; ,</span>{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; c</span>}}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; |</span>}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}{\sqrt{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; |</span>}{\mathrm{<span\; class="notranslate">\; e</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}\sqrt{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; |</span>}{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Wherein, n=1 or n=2,

if $\cos (\stackrel{\&Right\; Arrow;}{e},{\stackrel{\&Right\; Arrow;}{c}}_1)>\cos (\stackrel{\&Right\; Arrow;}{e},{\stackrel{\&Right\; Arrow;}{c}}_2)$ Then e is a valid email, otherwise e is an invalid email. At this point, we have a valid email for extracting user interests.

2. Effective Email Classification and User Interest Extraction

Each research field related to the team is divided into smaller sub-fields, and its background knowledge is represented by the prior knowledge set K _i of the sub-field n _i . K _i is a set of (n _k , a _k ), _nk is one of a group of keywords that can reflect the main content of nd _i together, a _k is the weight of _nk , indicating the ability of _nk to describe nd _i , The higher a _k is, the stronger the descriptive ability of _nk will be.

Classify valid emails by calculating the similarity between valid emails and the prior knowledge sets of each subfield, and use the subfields that members pay attention to to represent user interests according to the distribution of valid emails in each subfield;

First, calculate the probability that the content described by each valid email e refers to the subdomain n _i :

$\mathrm{<span\; class="notranslate">\; Sim</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; R</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}\mathrm{<span\; class="notranslate">\; f</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; kl</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; N</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

Among them, n _k is the subject of e-mail e and the keywords belonging to K _i contained in the text; D _l is the set of n _k , obviously, ( _nk , a _k ) ∈ _{K i} and _nk ∈ _{D l} ; S _kl is the number of occurrences of keyword n _k in the above part of email e and f(S _kl )=tanh(S _kl /3); R and N are the number of elements of _Dl and _Ki respectively.

Then, e is divided into the subfield with the highest probability to realize the classification of valid emails.

Generally speaking, most of the effective e-mails sent or received by users will be concentrated in a few subfields, and the research work they are engaged in should also be concentrated in these few subfields. That is to say, the subfields in the effective email collection are exactly the subfields that his research work focuses on, and the members' interest is represented by the set of the subfields they focus on. Therefore, based on the classification results of valid emails, the percentage of the user's research work involving each subfield can be calculated.

Then, calculate the percentage per _ij of user i's research work involving subfield j

${\mathrm{<span\; class="notranslate">\; per</span>}}_{\mathrm{<span\; class="notranslate">\; ij</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; nd</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&cap;</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; \&Element;</span>\mathrm{<span\; class="notranslate">\; from</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 2</span>}}^{<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; age</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; hl</span>}}}\mathrm{<span\; class="notranslate">\; sim</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; nd</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&cap;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; to</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 2</span>}}^{<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; age</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; hl</span>}}}\mathrm{<span\; class="notranslate">\; sim</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; from</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 2</span>}}^{<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; age</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; hl</span>}}}\mathrm{<span\; class="notranslate">\; sim</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&beta;</span>}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; \&Element;</span>{\mathrm{<span\; class="notranslate">\; to</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 2</span>}}^{<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; age</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; hl</span>}}}\mathrm{<span\; class="notranslate">\; sim</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; e</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; K</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; 100</span>}<span\; class="notranslate">\; \%</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

使得电子邮件的描述能力随其存在时间的增长而降低，age(e)是当前日期与电子邮件e的发送日期的差，hl＝30表明30天前的电子邮件只有当前电子邮件一半的描述能力；from_i是用户i所发送的有效电子邮件的集合，to_i是用户i所接收的有效电子邮件的集合。per _ij is the percentage of user i's research work involving subfield j, where from _i is the set of valid emails sent by user i, to _i is the set of valid emails received by user i; α=1, β = 0.8, which respectively represent the ability to describe the interests of the valid emails sent by the user and the valid emails received by the user,

The descriptive ability of email decreases with the increase of its existence time, age(e) is the difference between the current date and the sending date of email e, hl=30 indicates that the email 30 days ago has only half the description ability of the current email ; from _i is the collection of valid emails sent by user i, and to _i is the collection of valid emails received by user i.

The user's ability to receive valid e-mails from other members describing his interests depends on the sending member's knowledge of his research work; valid e-mails sent by users generally accurately reflect his research interests, Therefore, more descriptive power is given to effective emails sent by users. User research priorities tend to change over time, so e-mail should also become less descriptive as it ages, by adding Implemented by introducing a formula;

Finally, if per _ij is greater than the threshold, the subfield nd _j is added to the set of subfields concerned by user i, where the threshold is 10%.

3. Document understanding and classification

A basic concept, viewpoint or method is called a point of interest, and we use a Semantic Link Network (SG) to represent the semantic information of a point of interest. The interest point set SG- _{set i} of the subfield ndi describes all the semantics contained in _ndi , and its elements are the semantic link network corresponding to the interest points contained in _ndi . Divide the document into subfields with similar semantics by using the interest point set of the subfield as a template;

SG=(N, R), wherein, N is a collection of nodes, including an interest point N ₁ and a group of keywords {N ₂ , N ₃ ,..., N _m } that jointly represent the semantics of the interest point N ₁ ; R is a collection of directed arcs, representing causal relationships between nodes.

Figure 2(a) is a semantic chain network. The directed arc starting from N _i and ending at N _j represents the causal relationship from N _i to N _j , and its weight w _ij indicates the influence of the cause node N _i on the result node N _j degree, w _ij ∈ [-1, +1].

Figure 2(b) is the adjacency matrix representation of the semantic link network, which is an n×n matrix, and n is the number of nodes contained in the semantic link network. If there is a causal relationship between N _i and N _j , then the element in row i and column j of the adjacency matrix is w _ij , otherwise it is 0.

Figure 3 is a flowchart of document understanding and division. The specific steps are as follows:

S3-1. Select a document d from the team document database;

S3-2. Select a subfield nd _i to obtain the corresponding interest point set SG-set _i ;

S3-3. Calculate the semantic similarity md(d, nd _i ) between document d and subfield n _i :

S3-3.1 Divide the document d into several small parts: p ₁ , p ₂ , . . . , p _m , which can be divided by the number of bytes or paragraphs. Here it is divided by subsection, including subsections for further division;

S3-3.2 For any small part p _j , set md _Part-ji = 0;

S3-3.3 For any element SG _r of the interest point set SG-set _i in the subfield nd _i

(1) Calculate the state value V _k ′ of any keyword N _k included in SG _r in p _j : V _k ′=tanh(S _k /3), S _k is the number of times N _k appears in p _j ;

(2) V ₁ ′, V ₂ ′, ..., V _m ′)=(0, V ₂ ′, ..., V _m ′)×E _r , where E _r is the adjacency matrix representation of SG _r ;

(3) If md _Part-ji < V ₁ ′, then md _Part-ji = V ₁ ′

$S3-3.4\mathrm{md}(d,\mathrm{no}{d}_i)=\underset{j=1}{\overset{m}{\mathrm{\&Sigma;}}}m{d}_{\mathrm{part}-\mathrm{the\; ji}}/m,$ Among them, the document d is divided into m small parts

S3-4. If md(d, n _i )>0.65, divide document d into sub-field n _i , and go to S3-2.

The above method calculates the semantic similarity between each document and each sub-domain from the point of interest level, so that the document is classified into the sub-domain with higher semantic similarity, and a document may belong to multiple sub-domains at the same time. The team's document database not only includes a large number of existing technical documents, but also receives documents uploaded by team members, so as to continuously increase the capacity of the document database. Therefore, it is necessary to regularly check whether there are newly added documents in the document database, and if so, divide them into corresponding subfields according to the above method.

4. Effectively push documents according to user interests

The result of document understanding is to divide the documents in the team document database into various sub-domains related to the team. Write a document delivery program (eg, FileDeliver), which selects appropriate documents from the team's document database and pushes them to team members based on the document classification results and user interests. Because team members tend to read documents related to the subdomains they care about, the program pushes documents as email attachments to users whose subdomains of interest set includes the subdomain to which the document belongs.

Each document in the document database has two lists of "sent by" and "uploaded by". The "Sent People" list records which team members the document has been pushed to. When FileDeliver runs, it only pushes the document to the team members who do not appear in the "Sent People" list of the document. When a member uploads a document to the team's document database, if the document does not exist in the document database, the upload is successful, otherwise the prompt will be repeated. Regardless of whether the upload is successful or not, the member will be recorded in the "Uploaded by" list of the document. Because the document a member is trying to upload must be a document he already owns, FileDeliver will not push the document to members who already appear in the "uploaded by" list of the document. Team members only need to perform a simple upload operation to share the document among all members who need the document, which is simple and effective.

Claims (10)

1. The method for finding user interest in the E-mail stream and effectively pushing the document according to the user interest comprises the following steps of firstly, storing the E-mails among team members into an E-mail database and extracting effective E-mails from the E-mails; then, extracting user interests according to the distribution rule of the effective e-mails, and realizing classification of the documents in the team document database through semantic analysis; and finally, according to the user interest and the document classification result, pushing the document consistent with the member interest to the team members through the E-mails.

2. The method for discovering user interests in an email stream and efficiently pushing documents in accordance therewith as recited in claim 1, wherein the emails among team members are decoded by the email collecting program and the decoded contents are stored in the email database, and automatic warehousing of the emails is realized by periodically running the email collecting program, wherein spam is mostly from strange email addresses, and the process only considers the emails among the members, thereby eliminating the interference of the spam when extracting the user interests.

3. The method for discovering user interests in an email stream and for efficiently pushing documents as claimed in claim 1, wherein the method of natural language learning is used to obtain efficient emails that provide useful information for describing user interests, taking into account only the interests of team members in research and development work, thereby ensuring the accuracy of the extracted user interests.

4. The method of claim 1 for discovering user interests in an email stream and for efficiently pushing documents based thereon, further comprising subdividing a research domain associated with a team into sub-domains, establishing a priori knowledge sets of the sub-domains to represent background knowledge thereof, classifying the valid emails by similarity calculations between the valid emails and the prior knowledge sets of the sub-domains, and representing user interests by a set of sub-domains of interest to the members based on the distribution of the valid emails in the sub-domains.

5. The method of claim 1 for discovering user interests in an email stream and efficiently pushing documents in response thereto, wherein a time factor is introduced into the interest extraction process in view of possible changes in user interests over a longer period of time, the user interests are updated in time as new emails are generated and time passes, and pushing documents to users in response to user interests ensures that documents are always pushed to all team members who need the documents without either misposting or missed posting.

6. The method for discovering user interest in e-mail stream and effectively pushing documents according to the same as the claim 1, wherein an interest point set describing the sub-domain semantics is constructed, the documents are divided into sub-domains similar to the sub-domains of the sub-domains by taking the interest point set as a template, the documents are pushed to the members concerning the sub-domains to which the documents belong by a document pushing program, the documents pushed to the users are guaranteed to be needed by the users semantically, and the members of the team can complete the pushing of the documents by the program only by uploading the documents to a document database of the team, so that the method is simple and easy to implement.

7. A method for discovering user interest in an email stream and effectively pushing documents according to the user interest is characterized by mainly comprising the following four parts:

firstly, the e-mail is automatically stored, and the effective e-mail is extracted, wherein,

1. building an email database

The team members use the uniform E-mail server and server program to establish a database file under a certain directory of the E-mail server to store the E-mail information among the team members;

2. automatic e-mail storage

Firstly, all the e-mails among the team members are automatically forwarded to a fixed account by a mail server program, and the mails of the account are stored in a certain fixed directory of a mail server; then, regularly running the compiled mail collection program to realize the automatic storage of the e-mails, and decoding the e-mails by the program and storing the decoding result in the corresponding field of the e-mail database;

3. extracting valid emails

The invention only considers the interest of the user in the aspect of scientific research work, and extracts the effective e-mail which can provide useful information for describing the interest of the user through a natural language learning method;

second, efficient email classification and user interest extraction

Dividing the research fields related to the team intoSmaller sub-fields and through sub-fields nd_iPrior knowledge set K_iRepresenting its background knowledge, K_iIs (n)_k，a_k) Set of (2), n_kIs able to reflect nd together_iOne of a set of keywords of the primary content, a_kIs n_kWeight of (2) represents n_kTo nd_iDescription capability of a_kThe higher n_kThe stronger the description capability;

classifying the effective e-mails through similarity calculation of the effective e-mails and the prior knowledge sets of each sub-field, and expressing user interest by using the sub-field set concerned by members according to the distribution condition of the effective e-mails in each sub-field;

third, document understanding and classification

A basic concept, point of view or method is called a point of interest, and a semantic link network (SG) represents semantic information of a point of interest, where SG is (N, R), where N is a set of nodes, including a point of interest N₁And a group of points of interest N represented together₁Semantic keywords { N₂，N₃，...，N_m}; r is a set of directed arcs representing causal relationships between nodes, the sub-domain nd_iInterest point set SG-set_iDescription nd_iAll the implied semantic information, its elements are nd_iThe semantic chain network corresponding to the contained interest points divides the document into sub-fields with similar semantics by taking the interest point set of the sub-fields as a template;

fourthly, effectively pushing documents according to user interests

Writing a document pushing program, wherein the document pushing program pushes a document to a user of a concerned sub-field set including a sub-field to which the document belongs in the form of an e-mail attachment, each document has two lists of 'sent person' and 'uploading person', the document pushing program only pushes the document to team members which do not appear in the two lists, repeated sending is avoided, and the members can share the document among all members needing the document only by uploading the document to a team document database, so that the method is simple and effective.

8. The method for discovering user interests in an email stream and thereby efficiently pushing documents according to claim 7, wherein the first, email is automatically archived, the available emails are extracted, wherein,

3. extracting valid emails

First, a certain number of valid e-mails and invalid e-mails are selected as the training set C of valid e-mails respectively₁And invalid e-mail training set C₂And obtaining the standard vectors of valid e-mails and invalid e-mails by the following formulas

And

represents:

<math> <mrow> <msub> <mover> <mi>c</mi> <mo>&RightArrow;</mo> </mover> <mn>1</mn> </msub> <mo>=</mo> <mn>16</mn> <mfrac> <mn>1</mn> <mrow> <mo>|</mo> <msub> <mi>C</mi> <mn>1</mn> </msub> <mo>|</mo> </mrow> </mfrac> <munder> <mi>&Sigma;</mi> <mrow> <mi>e</mi> <mo>&Element;</mo> <msub> <mi>C</mi> <mn>1</mn> </msub> </mrow> </munder> <mfrac> <mover> <mi>e</mi> <mo>&RightArrow;</mo> </mover> <mrow> <mo>|</mo> <mover> <mi>e</mi> <mo>&RightArrow;</mo> </mover> <mo>|</mo> </mrow> </mfrac> <mo>-</mo> <mn>4</mn> <mfrac> <mn>1</mn> <mrow> <mo>|</mo> <msub> <mi>C</mi> <mn>2</mn> </msub> <mo>|</mo> </mrow> </mfrac> <munder> <mi>&Sigma;</mi> <mrow> <mi>e</mi> <mo>&Element;</mo> <msub> <mi>C</mi> <mn>2</mn> </msub> </mrow> </munder> <mfrac> <mover> <mi>e</mi> <mo>&RightArrow;</mo> </mover> <mrow> <mo>|</mo> <mover> <mi>e</mi> <mo>&RightArrow;</mo> </mover> <mo>|</mo> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>1</mn> <mo>)</mo> </mrow> </mrow> </math>

<math> <mrow> <msub> <mover> <mi>c</mi> <mo>&RightArrow;</mo> </mover> <mn>2</mn> </msub> <mo>=</mo> <mn>16</mn> <mfrac> <mn>1</mn> <mrow> <mo>|</mo> <msub> <mi>C</mi> <mn>2</mn> </msub> <mo>|</mo> </mrow> </mfrac> <munder> <mi>&Sigma;</mi> <mrow> <mi>e</mi> <mo>&Element;</mo> <msub> <mi>C</mi> <mn>2</mn> </msub> </mrow> </munder> <mfrac> <mover> <mi>e</mi> <mo>&RightArrow;</mo> </mover> <mrow> <mo>|</mo> <mover> <mi>e</mi> <mo>&RightArrow;</mo> </mover> <mo>|</mo> </mrow> </mfrac> <mo>-</mo> <mn>4</mn> <mfrac> <mn>1</mn> <mrow> <mo>|</mo> <msub> <mi>C</mi> <mn>1</mn> </msub> <mo>|</mo> </mrow> </mfrac> <munder> <mi>&Sigma;</mi> <mrow> <mi>e</mi> <mo>&Element;</mo> <msub> <mi>C</mi> <mn>1</mn> </msub> </mrow> </munder> <mfrac> <mover> <mi>e</mi> <mo>&RightArrow;</mo> </mover> <mrow> <mo>|</mo> <mover> <mi>e</mi> <mo>&RightArrow;</mo> </mover> <mo>|</mo> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>2</mn> <mo>)</mo> </mrow> </mrow> </math>

wherein, <math> <mrow> <mover> <mi>e</mi> <mo>&RightArrow;</mo> </mover> <mo>=</mo> <mrow> <mo>(</mo> <msub> <mi>e</mi> <mn>1</mn> </msub> <mo>,</mo> <msub> <mi>e</mi> <mn>2</mn> </msub> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>.</mo> <mo>,</mo> <msub> <mi>e</mi> <mrow> <mo>|</mo> <mi>F</mi> <mo>|</mo> </mrow> </msub> <mo>)</mo> </mrow> </mrow> </math> is a vector representation of an e-mail e, e_iIs the keyword w_iNumber of occurrences in the subject and body of e-mail;

is thatThe vector length of (d); i C₁I and I C₂Each is C₁And C₂The number of training samples, i.e., the number of electronic mail pieces contained,

then, the e-mail e is calculatedVector representation

And a standard vector

And

the calculation method of the similarity is as follows:

<math> <mrow> <mi>cos</mi> <mrow> <mo>(</mo> <mover> <mi>e</mi> <mo>&RightArrow;</mo> </mover> <mo>,</mo> <msub> <mover> <mi>c</mi> <mo>&RightArrow;</mo> </mover> <mi>n</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <mfrac> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mrow> <mo>|</mo> <mi>F</mi> <mo>|</mo> </mrow> </msubsup> <msub> <mi>e</mi> <mi>i</mi> </msub> <msub> <mi>c</mi> <mi>i</mi> </msub> </mrow> <mrow> <msqrt> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mrow> <mo>|</mo> <mi>F</mi> <mo>|</mo> </mrow> </msubsup> <msubsup> <mi>e</mi> <mi>i</mi> <mn>2</mn> </msubsup> </msqrt> <msqrt> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mrow> <mo>|</mo> <mi>F</mi> <mo>|</mo> </mrow> </msubsup> <msubsup> <mi>c</mi> <mi>i</mi> <mn>2</mn> </msubsup> </msqrt> </mrow> </mfrac> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>3</mn> <mo>)</mo> </mrow> </mrow> </math>

wherein n-1 or n-2,

if it is not <math> <mrow> <mi>cos</mi> <mrow> <mo>(</mo> <mover> <mi>e</mi> <mo>&RightArrow;</mo> </mover> <mo>,</mo> <msub> <mover> <mi>c</mi> <mo>&RightArrow;</mo> </mover> <mn>1</mn> </msub> <mo>)</mo> </mrow> <mo>></mo> <mi>cos</mi> <mrow> <mo>(</mo> <mover> <mi>e</mi> <mo>&RightArrow;</mo> </mover> <mo>,</mo> <msub> <mover> <mi>c</mi> <mo>&RightArrow;</mo> </mover> <mn>2</mn> </msub> <mo>)</mo> </mrow> </mrow> </math> E is a valid email, otherwise e is an invalid email.

9. The method for discovering user interest in an email stream and thereby efficiently pushing documents according to claim 7, wherein two, efficient email classification and user interest extraction, wherein,

first of all, what is described as a computationally efficient e-mail e relates to the sub-domain nd_iProbability of (c):

<math> <mrow> <mi>Sim</mi> <mrow> <mo>(</mo> <mi>e</mi> <mo>,</mo> <msub> <mi>K</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>k</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>R</mi> </munderover> <msub> <mi>&alpha;</mi> <mi>k</mi> </msub> <mi>f</mi> <mrow> <mo>(</mo> <msub> <mi>S</mi> <mi>kl</mi> </msub> <mo>)</mo> </mrow> <mo>/</mo> <mi>N</mi> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>4</mn> <mo>)</mo> </mrow> </mrow> </math>

wherein n is_kBelonging to K contained in the subject and body of e-mail_iThe keyword of (1); d_lIs n_kA set of (a); s_klIs the keyword n_kNumber of occurrences in the above-mentioned part of e-mail and f (S)_kl)＝tanh(S_kl3); r and N are each D_lAnd K_iThe number of elements (c). E is divided into the sub-fields with the highest probability to realize the classification of effective e-mails;

then, the research work to calculate user i involves the percentage per of the sub-domain j_ij

<math> <mrow> <msub> <mi>per</mi> <mi>ij</mi> </msub> <mo>=</mo> <mfrac> <mrow> <mi>&alpha;</mi> <msub> <mi>&Sigma;</mi> <mrow> <mrow> <mo>(</mo> <msub> <mrow> <mi>e</mi> <mo>&Element;</mo> <mi>nd</mi> </mrow> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>&cap;</mo> <mrow> <mo>(</mo> <mi>e</mi> <mo>&Element;</mo> <msub> <mi>from</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> </msub> <msup> <mn>2</mn> <mrow> <mo>-</mo> <mfrac> <mrow> <mi>age</mi> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> </mrow> <mi>hl</mi> </mfrac> </mrow> </msup> <mi>sim</mi> <mrow> <mo>(</mo> <mi>e</mi> <mo>,</mo> <msub> <mi>K</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>+</mo> <mi>&beta;</mi> <msub> <mi>&Sigma;</mi> <mrow> <mrow> <mo>(</mo> <mi>e</mi> <mo>&Element;</mo> <msub> <mi>nd</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>&cap;</mo> <mrow> <mo>(</mo> <mi>e</mi> <mo>&Element;</mo> <msub> <mi>to</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </mrow> </msub> <msup> <mn>2</mn> <mrow> <mo>-</mo> <mfrac> <mrow> <mi>age</mi> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> </mrow> <mi>hl</mi> </mfrac> </mrow> </msup> <mi>sim</mi> <mrow> <mo>(</mo> <mi>e</mi> <mo>,</mo> <msub> <mi>K</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> </mrow> <mrow> <mi>&alpha;</mi> <msub> <mi>&Sigma;</mi> <mrow> <mo>(</mo> <mi>e</mi> <mo>&Element;</mo> <msub> <mi>from</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </msub> <msup> <mn>2</mn> <mrow> <mo>-</mo> <mfrac> <mrow> <mi>age</mi> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> </mrow> <mi>hl</mi> </mfrac> </mrow> </msup> <mi>sim</mi> <mrow> <mo>(</mo> <mi>e</mi> <mo>,</mo> <msub> <mi>K</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> <mo>+</mo> <mi>&beta;</mi> <msub> <mi>&Sigma;</mi> <mrow> <mo>(</mo> <mi>e</mi> <mo>&Element;</mo> <msub> <mi>to</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> </msub> <msup> <mn>2</mn> <mrow> <mo>-</mo> <mfrac> <mrow> <mi>age</mi> <mrow> <mo>(</mo> <mi>e</mi> <mo>)</mo> </mrow> </mrow> <mi>hl</mi> </mfrac> </mrow> </msup> <mi>sim</mi> <mrow> <mo>(</mo> <mi>e</mi> <mo>,</mo> <msub> <mi>K</mi> <mi>j</mi> </msub> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>&times;</mo> <mn>100</mn> <mo>%</mo> <mo>-</mo> <mo>-</mo> <mo>-</mo> <mrow> <mo>(</mo> <mn>5</mn> <mo>)</mo> </mrow> </mrow> </math>

Wherein α ═ 1, β ═ 0.8, represent the description ability of interest to the valid email sent by the user and the valid email received, respectively;making the description capacity of the e-mail reduce with the increase of the existing time, wherein, age (e) is the difference between the current date and the sending date of the e-mail, and hl is 30, which indicates that the e-mail before 30 days has the description capacity of half of the current e-mail; from_iIs the set of valid e-mails, to, sent by user i_iIs the set of valid emails received by user i, if per_ijGreater than threshold value, and sub-domain nd_jJoin the set of sub-domains of interest to user i, where the threshold is 10%.

10. The method for discovering user interests in an email stream and pushing documents efficiently according to the same as claimed in claim 7, wherein the third step of document understanding and dividing is as follows:

s3-1, selecting a document d from the team document database;

s3-2, selecting a sub-field nd_iObtaining the point of interest set SG-set_i；

S3-3, calculating the document d and the sub-field nd_iSemantic matching degree md (d, nd)_i)：

S3-3.1 divides document d into several small parts: p is a radical of₁，p₂，…，p_m；

S3-3.2 for any small part p_jLet md be_Part-ji＝0；

S3-3.3 subfield nd_iInterest point set SG-set_iAny element SG_r

(1) Calculate SG_rAny keyword N contained_kAt p_jState value V in_k′:V_k′＝tanh(S_k/3)，S_kIs N_kAt p_jThe number of occurrences in (a);

(2)(V₁′，V₂′，...，V_m′)＝(0，V₂′，...，V_m′)×E_r，E_ris SG_rA adjacency matrix representation of (a);

(3) if md is_Part-ji＜V₁Then md_Part-ji＝V₁′

S3-3.4 <math> <mrow> <mi>md</mi> <mrow> <mo>(</mo> <mi>d</mi> <mo>,</mo> <msub> <mi>nd</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>=</mo> <munderover> <mi>&Sigma;</mi> <mrow> <mi>j</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>m</mi> </munderover> <msub> <mi>md</mi> <mrow> <mi>Part</mi> <mo>-</mo> <mi>ji</mi> </mrow> </msub> <mo>/</mo> <mi>m</mi> <mo>,</mo> </mrow> </math> Wherein the document d is divided into m small parts

S3-4. if md (d, nd)_i) > 0.65, partition document d into sub-domains nd_iGo to S3-2.

Images