WO2014023121A1

WO2014023121A1 - Method and device for launching individual content

Info

Publication number: WO2014023121A1
Application number: PCT/CN2013/076308
Authority: WO
Inventors: 游源; 钟杰萍; 尹攀; 杜家春
Original assignee: 华为技术有限公司
Priority date: 2012-08-10
Filing date: 2013-05-28
Publication date: 2014-02-13
Also published as: CN103577504A

Abstract

Provided is a method for launching individual content, comprising: behavioral targeting(BT) server receiving the user information sent by terminals, analyzing the web pages accessed by the terminals, and getting the types of the web pages(S101); according to the types of web pages and user information, obtaining the characteristic data of users(S102); querying the mapping table of user characteristic data and launching strategy data which is preconfigured by the BT server, and determining the corresponding launching strategy data for the user characteristic data(S103); writing the user characteristic data and corresponding launching strategy data into the cookies of the terminals, so that after receiving the connecting request from the terminals, content providing server can launch the content individually to the terminals in line with the launching strategy data in the cookies of the terminals(S104). Accordingly, present solution also provides an individual launching device, which reduces bandwidth consumption and improves the efficiency of server interaction.

Description

Method and device for personalizing content

This application claims priority to Chinese Patent Application No. 201210284928.8, entitled "A Method and Apparatus for Distributing Personalized Content", filed on August 10, 2012, the entire contents of which are incorporated by reference. In this application.

Technical field

Embodiments of the present invention relate to the field of communications, and in particular, to a method and apparatus for delivering personalized content. Background technique

As the pace of social informatization accelerates, new technologies and new media continue to emerge, and the content of Internet products exceeds demand, and the homogenization trend is obvious. The focus of the competition will shift to the quality of service, which is to provide services that meet the needs of customers. Personalized service is a concrete embodiment of customer satisfaction in the philosophical field. It embodies the business philosophy of “people-oriented” and is an important way for modern enterprises to improve their core competitiveness. How to collect and analyze specific user behaviors, providing personalized information and content to end users at low cost and high efficiency, and improving service quality are the hotspots of current industry research.

In the prior art, when a user accesses a content provider of a specific content, the provider sends the current user identification information to the analysis platform in real time, and the analysis platform acquires the current user feature from the database to which it belongs, and returns to the specific content. Business or third-party content providers to provide personalized content. For example, when the user logs in to the website, the website needs to communicate with the user behavior analysis system in real time, and then the system retrieves the relevant user data information in real time, and returns the current user type result to the current website (or third party advertisement provided in real time). The organization organizes the personalized content matching with the user and finally returns it to the current user terminal. The above steps include multiple real-time communication between servers. There are many intermediate links. The response speed and delivery reliability depend heavily on the user network environment. The bandwidth consumption is large and the server interaction efficiency is not high.

Summary of the invention

In view of this, the embodiments of the present invention provide a method and an apparatus for delivering personalized content, which solves the problem of large bandwidth consumption and low server interaction efficiency when customizing personalized content in the prior art.

The embodiment of the invention provides a method for delivering personalized content, including:

The user behavior analysis BT server obtains the user information sent by the terminal during the process of accessing the webpage, analyzes the webpage accessed by the terminal, and obtains the webpage type of the webpage; and obtains according to the webpage type of the webpage and the user information. User characteristic data; Querying a correspondence table between the user feature data pre-configured by the BT server and the delivery policy data, and determining the delivery policy data corresponding to the user feature data;

The user feature data and the corresponding delivery policy data are written into the cookie of the terminal, so that the content providing server receives the access request sent by the terminal when the next time the webpage is accessed by the terminal, according to the embodiment of the present invention. A BT server that delivers personalized content, including:

An obtaining unit, configured to acquire user information sent by the terminal during the process of accessing the webpage, and analyze the webpage accessed by the terminal to obtain a webpage type of the webpage;

The obtaining unit is further configured to acquire user feature data according to the webpage type of the webpage and the user information;

a determining unit, configured to query a correspondence table between the user feature data pre-configured by the BT server and the delivery policy data, and determine the delivery policy data corresponding to the user feature data;

a writing unit, configured to write the user feature data and the corresponding delivery policy data into the cookie of the terminal, so that the content providing server receives the access request sent by the terminal when the next time the webpage is accessed, according to the terminal The delivery policy data in the cookie is personalized to the terminal.

In the embodiment of the present invention, after receiving the user information sent by the terminal, the BT server performs webpage type analysis on the webpage accessed by the terminal, and obtains the user characteristic data and the corresponding delivery policy according to the user information and the webpage type, and writes the terminal to the terminal. The cookie, so that the next time the content providing server receives the terminal access request, the terminal can directly deliver the personalized content to the terminal by reading the cookie of the terminal. Reduces bandwidth consumption and improves server interaction efficiency.

DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the prior art, the drawings used in the embodiments or the description of the prior art will be briefly described below. Obviously, the drawings in the following description It is a certain embodiment of the present invention, and other drawings can be obtained from those skilled in the art without any creative work.

1 is a system architecture diagram of an embodiment of the present invention;

2 is a flowchart of Embodiment 1 of the present invention;

3 is a flowchart of Embodiment 2 of the present invention; FIG. 4 is a structural diagram of Embodiment 3 of the present invention.

detailed description

Figure 1 shows the architecture of the system. As shown in FIG. 1, the system includes a terminal, a content providing server, and a Behavioral Targeting (BT) server. The system communicates with each other via a wired or wireless communication network. These communication networks include, but are not limited to, a Mobile Telephone Network, a Wireless Local Area Network (LAN), a Bluetooth Personal Area Network, an Ethernet LAN, and a Token Ring. A token ring LAN, a wide area network, the Internet, and so on. The terminal may include, but is not limited to, a mobile device, a combination PDA and mobile telephone, a PDA, an integrated information device (IMD), a personal computer (PC). And a notebook computer (Notebook Computer). These terminals can be moved or located on a mobile device such as, but not limited to, a car, a truck, a taxi, a bus, a ship, an airplane, a bicycle, a motorcycle, and the like.

The above communication device can implement communication process based on various different transmission technologies, including but not limited to Code Division Multiple Access (CDMA), Global System for Mobile Communications (GSM), Universal Mobile Communication. System (Universal Mobile Telecommunications System, UMTS), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Transmission Control Protocol/Internet Protocol (Transmission Control Protocol/Internet Protocol, TCP/IP), Short Messaging Service (SMS), Multimedia Messaging Service (MMS), e-mail, Instant Messaging Service (IMS), Bluetooth, IEEE 802.11, etc. Different media resources may be used between the above communication devices, including but not limited to, radio, infrared, laser, cable connection, and the like.

The content providing server registers its service with the BT server in advance, and embeds a specific script pointing to the BT server into the web page it operates, thereby forming a monitoring domain of the BT server. When the user accesses the page in the above monitoring domain through the terminal, the page will automatically load and execute the above script file. The script file can automatically collect and store the information that the user is browsing on the page and send it to the BT server. The BT server analyzes and processes the specific information of the user in the entire monitoring domain according to the relevant model algorithm, and forms user characteristic data corresponding to the current user, and according to the content providing server, the server is configured in advance for the user characteristic data in the BT server. The personalized personalized content delivery rules form the corresponding strategy. The above information is written to the user's local cookie in the agreed format through the above specific script. Therefore, when the user visits the website again, the content providing server can directly acquire and execute the BT analysis result and the personalized delivery policy data stored in the user locality, and perform accurate delivery. That is, the BT server only needs to analyze the user once. When the user visits the website again, the content providing server directly feeds the personalized content to the end user by reading the local cookie of the terminal, and does not need to re-click the stream data through the BT server. Analysis is performed to reduce bandwidth consumption and improve server interaction efficiency.

Figure 2 is a flow chart for delivering personalized content. As shown in Figure 2, the figure includes:

S101, the user behavior analysis, the BT server acquires the user information sent by the terminal during the process of accessing the webpage, and analyzes the webpage accessed by the terminal, and obtains the webpage type of the webpage;

After the terminal accesses the webpage provided by the content providing server, the terminal executes the script program on the webpage through the browser, and sends the user information to the BT server. User information includes HTTP request information, such as requested URL, jump source URL, etc., information of the requested page, such as page title, keywords, abstracts, etc., and user behavior information, such as click, submit, input, jump, Refresh and so on.

The BT server can perform checksum reconstruction on the received user information. The user information for successful verification needs to be reconstructed into click stream data, and the user information for verification failure is deleted, and new user information is re-received and verified.

The BT server obtains the user information, and also needs to analyze the webpage accessed by the terminal to obtain the webpage type of the webpage. It should be noted that the steps for the BT server to obtain user information and analyze the web pages accessed by the terminal are not strictly time-limited, and may be performed simultaneously, or the user information may be acquired first, and then the accessed web pages may be analyzed.

The BT server analyzes the webpage accessed by the terminal, and the specific analysis steps are as follows: Set the type set of the webpage and the set of frequencies corresponding to the type of the webpage.

The BT server divides the content of the webpage into N categories in advance, that is, sports, finance, science and technology, etc., and is represented by {ς ₂ ,...,^}. And set the collection

} , where Μ,. is the frequency of the type. The frequency is the number of web pages of the type in the total number of web pages accessed by the terminal. It should be noted that {ς ₂ ,...,^} And {MM^.^M are both set in advance by the BT server, which is different from the webpage type of the webpage actually accessed by the terminal. The latter requires the BT server to pass {ς, ί^..., ^} and {MA^ .^Mw} and the feature data of the web page accessed by the terminal are calculated.

Obtaining feature data of the webpage, where the feature data includes key terms, character spacings, and text lengths corresponding to webpage types in the type set of the webpage.

Calculating a probability of the feature data according to a set of frequencies corresponding to the type of the webpage, and selecting a maximum one or more probability values of the calculated probability values, and acquiring a webpage type corresponding to the selected probability value.

The calculation method is as follows: The BT server calculates each prioritized ςΕ{ς ₂ ,...,^}, and calculates its prior probability 7^,.) = ^ and corpse (=^, where Μ = 5 1. is the training set Total number of texts, indicating input

Μ Μ ^{1 The} data does not belong to the number of classification G. Calculate the conditional probability p(F _i =x _k \c _l )= ^画^ := Q+i of the feature G in the classification G, where _COM „^ _=3⁄4 , _c .) denotes the classification _c . take

^ count(F _i = ^, .) + 1 F _f | The number of times the value indicates the number of values that the feature can take. For the information text according to Bayesian principle and Naive Bayes hypothesis, the probability that the information text is classified as G is proportional to a _t

, where (d) represents the value of the feature in the message text. Therefore, the BT server determines the value of " ^ Qn^^^ ^ IQ, and expresses the result as a set V={a^..., a _N } ^ The BT server can sort P by its element value. , as in descending order, get an ordered collection

Q, that is, the larger the value of the element, the higher the position. The BT server may determine that the text belongs to the category corresponding to the element ranked first in the set Q, and may also select the elements of the top f as the probability distribution of the corresponding category, and the selection rule is pre-determined by the BT server. Configuration. For example, CI and C2 respectively represent sports and tourism. The text d of the web page to be analyzed contains feature data such as "basketball", "soccer", "travel", etc., and the ordered set Q obtained after the above operation is obtained. In the first place The element corresponds to CI, and the second element corresponds to C2. If only the first element is selected, the BT server determines that the type of the text d of the webpage is of the type C1, that is, the sports class, if the first and the first are selected The two-digit element determines that the type of the text d of the webpage is both a sports class and a ^game class.

S102. Acquire user feature data according to the webpage type of the webpage and the user information. The method for specifically acquiring user feature data is as follows:

Calculating a word frequency TF, an inverse document frequency IDF of the metadata corresponding to the webpage type, and a product TF-IDF of the TF and the IDF according to the click stream data;

The BT server can calculate the word frequency by the Markov model formula. The formula is as follows:

Where ^s is the set of historical effective steps, E a _M } is a metadata in the metadata set, and the metadata is data representing key terms of the web page type. If the webpage type is sports, the metadata may be " Sports ", or "soccer", or "basketball", etc., whose metadata is preset by the BT server. The webpage type P accessed by the terminal is the target webpage in the Markov process, and the target webpage is the webpage accessed by the user in a time period, and f(k) is a decaying attenuation factor for expressing the previous step, for example, this embodiment Where f(k) = e-Pkt , the smaller the value of the attenuation factor with time, that is, the smaller the weight, the lower the reference value for the user behavior for the user behavior. P ( xk = ai|xO = p ) indicates the frequency at which ai appears in a target web page, that is, the ratio of the number of occurrences of the word frequency ai in the target web page p to the total number of occurrences of all words of the target web page p, tfai,p indicates Term Frequency (TF), the word frequency indicates how often an entry appears in a document or landing page. The formula indicates the frequency at which a particular entry appears on a web page that the user has visited from the past to the present. For example, to calculate the word frequency of the word "basketball" in 100 web pages accessed by the user from the past to the present, ΒΤ server based The above formula, through the input parameters: al = "basketball", S = 100, and counts P (xk

= ai|xO = p ), the word frequency of the word "basketball" can be calculated by the above formula.

After the BT server calculates the word frequency, you also need to bring tfai,p into the formula:

Tfid = if x ki. , where _{tfl J} = _{tfai p} .

For the above formula, you need to first TF-IDF (term frequency-inverse document Frequency ) To illustrate, TF-IDF is a commonly used weighting technique for retrieval and text mining. It is mainly used to evaluate the importance of a word for one of a file set. If a word or phrase appears frequently in an article and rarely appears in other articles, the word or phrase is considered to have a good class distinguishing ability and is suitable for classification.

Term Frequency (TF) refers to the frequency at which a given word appears in the file or landing page. In the embodiment of the present invention, the target webpage represents a webpage accessed by the terminal. Inverse Document Frequency (IDF) is a measure of the universal importance of a word. The principle is: If the document or landing page containing a certain entry has fewer IDF, the entry has a good class distinguishing ability. The IDF of a particular word can be obtained by dividing the total number of files by the logarithm of the resulting quotient:

Where |D|: the total number of files in the database or the total number of landing pages, IU: 3⁄4 ^ ⁽ 1 ί: the number of files containing words or the number of landing pages containing words, (ie the number of files in ^≠) if the word is not In the database or landing page, the divisor is zero, so it is generally used. Therefore, for the formula ¹ ^^^ = ^{tf x ic} , the high word frequency within a particular file is known, and the word is The low file frequency in the entire file set, multiplied to find high weight

TF-IDF. TF-IDF means filtering out common words and retaining important words. The higher the value, the more important the word is.

The BT server calculates the values of tfai,p and idf by the above formula, and multiplies the two to find the value of tfidf.

Selecting one or more TF-IDFs with the largest value, determining the metadata corresponding to the one or more TF-IDF values, querying the correspondence table between the metadata and the user feature data, and acquiring the user corresponding to the metadata Feature data.

For example, the BT server first obtains all types of web pages accessed by the terminal, including four specific types: "sports class", "music class", "financial class", and "skull class", and then select "sports", "music", "Finance", "IT" four specific terms, respectively, find the value of their tfidf, and then sorted from large to small, the BT server selects the largest one or the first few, assuming the order is " Sports, "music", "Financial", "IT", then the BT server determines that the user's user characteristics are "sports fans", or the user characteristics are "sports fans" and "music talents". The specific determination process can be configured by querying metadata and A correspondence table of user characteristics is obtained. The user feature may be represented by a string or a data format commonly used by other computers, and the embodiment is not limited.

S103: Query a correspondence table between user feature data and delivery policy data pre-configured by the BT server, and determine delivery policy data corresponding to the user feature data.

It should be noted that the correspondence table is pre-configured by the BT server. For example, a user's user characteristics are "ordinary white-collar/influx/gourmet", and the corresponding delivery strategy is "consumer electronic/meal-escaped characters, integer numbers, and other computer-used data format representations, which are represented by strings. They are "white_ collar /fashion- follower/ food-lover" and "electronic- consumption/ group_purchasing_of_food".

S104. Write the user feature data and the corresponding delivery policy data to the cookie of the terminal, so that the content providing server receives the access request sent by the terminal when the next time the webpage is accessed, according to the need, the terminal is The web page accessed at one time can be the web page visited in the previous step, or it can be another web page.

The cookie written is as shown in Table 1:

Table 1 Variables 1 UserTypes

White- collar

Variable 1 value I fashion— follower/

Food lover

Variable 2 Strategies

Electronic— consumpti

Variable 2 value on/

Group_purchasing_of_food

Domain www. example . com Optional sign 1536

Expiration time 2011-10-01 00:00:00

Creation time 2011-08-25 08:02:21

Among them, variable 1 and variable 2 represent user characteristic data and corresponding policies for the feature, respectively, and the remaining parameters, such as the domain, optional flag, expiration time and creation time, are standard parameters of the file format of the cookie. Thus, during the validity period of the above-mentioned cookie, the website can be personalized and pushed according to the relevant information in the cookie.

After the cookie is written, when the terminal requests to access the webpage provided by the content providing server again, the content providing server directly reads the cookie of the terminal, obtains the current user characteristic information, and directly specifies the delivery policy data content providing server to execute the personalized personalized policy data. A personalized content jump link is generated directly for the current user, or a personalized content jump link is requested from the third-party content server. When the terminal accesses the link, the content providing server delivers the personalized content to the user, and completes the accurate delivery.

Figure 3 is a method flow for delivering personalized content. As shown in Figure 3, the method flow includes:

S201, the BT server and the content providing server complete the registration process;

The content providing server requests the BT service, and carries the basic information of the website, such as the domain name, the basic service type, etc., in the request message, for example, the domain name is www.example.com, and the basic service type is the search type. After receiving the request from the content providing server, the BT server completes the registration process with the content providing server. After the registration, the BT server specifies a plurality of personalized delivery policies for one or more user characteristics according to the pre-configured user characteristics and personalized delivery policy correspondence table of the content providing server within the specified user feature set, and periodically The statistics of the number of users in the database, the corresponding user characteristics, and the delivery policy are presented to the content providing server in the form of a report. E.g,

The user feature set given by the BT server includes: ordinary white-collar workers, hipsters, gourmets, friends, chasing stars, etc., and the content providing server can perform any combination in the set to form a customized user group, such as defining a group 1 = {Ordinary white-collar workers and friends} (that is, both "ordinary white-collar workers" and "驴" "Users' characteristics of the crowd", group 2 = {潮人OR star chasing family} (that is, a specific group of people with "popular people" and "chasing stars" user characteristics), and then can be assigned to group 1" Personalized content for "safe outdoor activities", and the personalized content of "pop music and star dynamics" is assigned to group 2.

S202. The BT server collects user information.

When the user accesses the webpage containing the specific script of the BT server through the terminal, the BT server starts to automatically collect the current user information, including but not limited to: HTTP request information, such as the requested URL, the jump source URL, etc., is requested. Information about the page, such as page titles, keywords, abstracts, etc., as well as user behavior information such as clicks, submissions, inputs, jumps, refreshes, and more. The collection process may be: When the user accesses the webpage through the terminal, the terminal executes the command statement in the specific script and sends the user information to the remote BT server.

S203. Perform checksum reconstruction on user information.

Common errors or exception records include garbled data, duplicates, abends, and so on. The BT server will discard the above errors and abnormal data.

The BT server receives the above-mentioned user information and verifies it. There are many ways to check the BT server. For example, the format in which user information is saved is separated by some feature separators, such as "xxxxx ^Λ y yyyyy", where " ^Λ Β" is the feature separator, "x", "y" is the original data, and its form can be For characters or other forms of data commonly used in the computer field. If the format received by the device is not the above format, if the feature break or the feature break is not correct, the BT server determines that the data is abnormal, and performs abnormal and error culling, that is, discards the received abnormal user information. The data, waiting again and receiving data for the next set of user information. If the user information is not abnormal, the BT server reconstructs the user information into click stream data. Refactoring means reprocessing and integrating the data of user information. Specifically, the BT server converts the verified data into click stream data, and the type of the click stream data includes but is not limited to: a user browsing time period, a browsing time, a frequency of browsing the website, and a type of browsing the website. The browsing time period can be divided into morning (5:01 - 12:00), afternoon (12:01 - 6:00), evening (6:01 - 22:00) and late night (22: 01 - 5:00). The browsing duration indicates the length of time from the time the page request is initiated to the page closing time in the original data, including: short (less than 10 seconds;), normal (10-30 seconds;), longer (30 seconds to 100 seconds;), long ( More than 100 seconds;). It should be noted that the time period of the division is not strictly limited, and the remaining divisions also belong to the present. The scope of the invention is protected. The frequency of browsing the website indicates the number of visits to the website in a unit of time. The types of websites that can be browsed can be classified into sports, finance, and search. For example, the time to initiate a page request in the original data is 23:09:23, and the page close time is 23:11:01, then the above data can be reconstructed as: Access time period: Late night (22:01-05:00) Visit duration: Longer (30 seconds - 100 seconds); Visits: 1 time / day; Visit website type: Sports.

S204: analyzing the webpage in real time or at a time;

The BT server analyzes the content of the webpage in real time or periodically, and stores the webpage content analysis result in the database. The result is the type of the webpage, such as sports, finance, science, and the like.

Analysis can draw on commonly used training algorithms such as support vector machines (SVM), decision trees, neural networks, and Naive Bayes. First, the content of the website is classified, such as sports, finance, science and technology, education, military, etc., and then the above algorithm is used to train a classifier for the above website, and at the same time, a specific webpage is captured, and the content text of the website is put into the classifier for analysis. And give the corresponding content classification. Taking the naive Bayesian method as an example, the BT server calculates the prior probability ρ(ς)=^^ for each class ς e {C _X , C ₂ ,..., C _N }. ρ( = , where Μ=5 τι . is the total number of training set texts,

Μ Μ ¹ ^ indicates the number of input data that does not belong to the classification G. Calculate the conditional probability p( ΐς) = _m ^count(F ' ^=xc ') ⁺¹ of the feature value in the classification G, where _COM «t = _3⁄4 , Q denotes the classification c' ^

^ count(F _i = ^, .) + 1 F _f | The number of times the value is 1, 1 1 indicates the number of values that the feature can take. For the information text according to the Bayesian principle and the naive Bayesian hypothesis, the probability that the information text is classified as G is proportional to a _t = P(C _t ) · Π Ρ ( = ( ) IG ) , where ^d) represents information The value of the feature ^Fi in the text. Therefore, the BT server finds the value of " , .=^0;]^^= ( ^.), and expresses the result as a set P BT server can sort P by its element value, such as descending order, to get an ordered set.

Q, that is, the larger the value of the element, the higher the position. The BT server can determine that the text belongs to the set Q The classification corresponding to the element ranked first may also select the elements ranked in the first f as the probability distribution of the corresponding classification, and the selection rules are pre-configured by the BT server. For example, CI and C2 respectively represent sports and tourism. The text d of the web page to be analyzed contains feature data such as "basketball", "soccer", "travel", etc., and the ordered set Q obtained after the above operation is obtained. In the first place, the element corresponds to C1, and the second element corresponds to C2. If only the first element is selected, the BT server determines that the type of the text d of the webpage is of the type C1, that is, the sports class. If the first and second elements are selected, it is determined that the type of the text d of the webpage is both a sports class and a ^game class. S205. Acquire feature data of the user.

Under certain conditions, if the amount of information of a particular user has reached the preset value, or the predetermined cookie update period is reached, the BT server extracts the click stream data of the user in the database and the type of the visited webpage, and performs modeling analysis. Obtaining the characteristic data of the user, for example, the user 1 is both "ordinary white-collar" and "friend".

For the obtained user behavior, the Markov model or the probability model can be used to fit and analyze the relationship between the sequential behavior sequences, and an attenuation factor is added to express the decay of the previous step, that is, the longer the weight of the record is, the longer the weight is. small. Or use Bayesian estimation to model the user's click record type.

Take the Markov model as an example:

For the Markov model, the main idea is to estimate the user's current behavior of the web page from the user's past behaviors on the web page, that is, the relevance of the past behavior to the current behavior. Its formula is as follows:

s

Tf _ai , _P = ^ f(k)P(x _k = ai |x ₀ = p)

k=i

Where ^s is the set history effective number of steps, ai e {a^ .^ aM} is a metadata in the metadata set, and the metadata is data representing key terms of the web page type. The webpage type P accessed by the terminal is the target webpage in the Markov process, the target webpage is the webpage accessed by the user in a time period, and f(k) is a decaying attenuation factor for expressing the previous step, for example f(k) ) = e-Pkt, the smaller the value of the attenuation factor over time, that is, the smaller the weight, for the user behavior, it can be understood as the time The lower the user's behavior, the lower the reference value. P ( xk = ai|xO = p ) indicates the frequency at which ai appears in a target web page, that is, the ratio of the number of occurrences of the word frequency ai in the target web page p to the total number of occurrences of all words of the target web page p, tfai,p indicates Term Frequency (TF), the word frequency indicates how often the term appears in the document or landing page. The formula indicates the frequency at which a particular entry appears on a web page that the user has visited from the past to the present. For example, to calculate the word frequency of the word "basketball" in 100 web pages accessed by the user from the past to the present, the server is based on The above formula, through the input parameters: al=

"Basketball", S=100, and count the size of P ( xk = ai|xO = p ), and finally calculate the word frequency of the word "basketball".

t ^fid = ^{x id} , where tfij = tfki,p.

For the above formula, TF-IDF (term frequency-inverse document frequency) needs to be described first. TF-IDF is a commonly used weighting technique for retrieval and text mining. It is mainly used to evaluate a word for a file set. The importance of one of the documents. If a word or phrase appears frequently in an article and rarely appears in other articles, the word or phrase is considered to have a good class distinguishing ability and is suitable for classification.

The term frequency (TF) refers to the frequency at which a given word appears in the file. Inverse Document Frequency (IDF) is a measure of the general importance of a word. The principle is: If the document or landing page containing a certain term has fewer, the larger the IDF, the better the classification ability of the term. The IDF of a particular word can be obtained by dividing the total number of files by the number of files containing the word, and then taking the logarithm of the resulting quotient: id = log

{i: 3⁄4 G dy }

Where |D|: the total number of files in the database or the total number of landing pages, Ιϋ the number of files containing words or the number of landing pages containing words, (ie ^3⁄4 Γ≠ ^{: ί)} the number of files if the word is not in the database or In the landing page, it will cause the dividend to be zero, so it is generally used.

1 + |{i: k e f3⁄4}|.

Therefore, for the formula = ^tf y X ^k , the high word frequency within a particular file is known, and the low file frequency of the word in the entire file set is multiplied to find a high weight.

TF-IDF. TF-IDF means filtering out common words, retaining important words, the higher the value, the The more important the words are.

The BT server calculates the values of tfai,p and idf by the above formula, and multiplies the two to find the value of tfidf. For example, the BT server performs modeling and analysis on a certain user. The steps are as follows: Firstly, all types of web pages browsed by the user are obtained, and there are four specific categories of "sports class", "music class", "financial class", and "IT class". Type, then select "sports", "music", "finance", "IT" four specific terms, find the value of tfidf, and then arrange from big to small, BT server selects the largest one or In the first few, assuming that the order is "sports", "music", "finance", "IT", then the BT server determines that the user's user characteristics are "sports fans", or the user characteristics are "sports fans". "and "music talent", the specific determination process can be obtained by querying the configuration table of the terms and user characteristics. The user feature may be represented by a string or a data format commonly used by other computers, and the embodiment is not limited.

S206. Write user characteristic data and corresponding delivery policy data into the cookie.

The BT server matches the obtained user feature with the personalized delivery policy set by the content providing server that provides the webpage, obtains personalized delivery policy data for the user feature, and converts the format into a specified cookie format, and The included user characteristic data and the personalized delivery policy data are written into the cookie of the terminal, and the life cycle of the cookie, that is, the user characteristic data and the validity period of the corresponding delivery policy data, is set.

The embodiment of the present invention has no limitation on the manner of writing a cookie, and the writing method of forming a single cookie record under the current website domain name is taken as an example, and the user characteristic of the user is "ordinary white-collar/influx/gourmet". The corresponding delivery strategy is "consumer electronics/catering group purchase", and the written user characteristic data and corresponding delivery strategy data can be represented by a commonly used data format of a string, an escape character, an integer number, etc., where a string is used. , respectively, "white_collar/fashion-follower-following-food-lover", "electronic-consumption/group_purchasing_of_food" ^ The network accounts for www.example.com in the BT server monitoring domain, and the results are shown in Table 1.

S207. The terminal initiates an access request to the content providing server.

It should be noted that after the BT server writes the user feature data and the corresponding delivery policy data into the cookie, the user accesses the webpage again through the terminal.

S208. The content providing server acquires current user feature data and corresponding delivery policy data. The content providing server obtains current user feature information by reading the cookie information of the terminal. Its specified delivery strategy data.

S209. The content providing server delivers the personalized content to the user.

The content providing server directly executes the personalized delivery policy data, generates a personalized content jump link for the current user, or requests a personalized content jump link to the third-party content server. When the terminal accesses the link, the content providing server delivers the content to the user. Personalize the content and complete the precise delivery.

FIG. 4 is a structural diagram of a device of the BT server, as shown in FIG. 4, including:

The obtaining unit 301 is configured to acquire user information sent by the terminal during the process of accessing the webpage, and analyze the webpage accessed by the terminal, and obtain the webpage type of the webpage; after the terminal accesses the webpage provided by the content providing server, the terminal passes The browser executes the script program on the web page and transmits the user information to the obtaining unit 301. User information includes HTTP request information, such as requested URL, jump source URL, etc., information of the requested page, such as page title, keyword, abstract, etc., and user behavior information, such as click, submit, input, jump, Refresh and so on.

The obtaining unit 301 can perform checksum reconstruction on the received user information. User information for successful verification needs to be reconstructed into click stream data, and the user information for verification failure is deleted, and new user information is re-received and verified.

The obtaining unit 301 obtains the user information, and also needs to analyze the webpage accessed by the terminal to obtain the webpage type of the webpage. It should be noted that the step of obtaining the user information and analyzing the webpage accessed by the terminal by the obtaining unit 301 is not strictly limited, and may be performed simultaneously, or the user information may be acquired first, and then the accessed webpage may be analyzed. The BT server divides the content of the webpage into N categories in advance through the setting unit, that is, sports, finance, science and technology, etc., and is represented by {G'C^'Cw}. And set the set ^{^{^{{Μι 'Μ2' ··· 'Μ}}} ^, where G is a type of frequency. The frequency is the total number of web pages accessed by the terminal, ^C , the number of web pages of the type. Incidentally, {^ '^} and {^' ^ '^} _ΒΤ server are set in advance, and the type of web page different from actual access terminal, which is needed by server ΒΤ ^{CI ' ^C2 -' ^Cw ^ ^Ml ' ^M2 -' ^M w} and the feature data calculation of the web page accessed by the terminal And got it.

The obtaining unit 301 acquires feature data of the webpage, where the feature data includes a key term, a character pitch, and a text length corresponding to the webpage type in the type set of the webpage.

The calculating unit calculates a probability of the feature data according to a set of frequencies corresponding to the type of the webpage, and selects one or more probability values of the calculated probability values to obtain a webpage corresponding to the selected probability value. Types of.

The calculation method is as follows: BT server's calculation unit for each classification

Calculate its prior probability ^^^ and ^^^, where Μ = 5 1. is the total number of texts in the training set.

MM ¹ indicates the number of input data that does not belong to the classification G. Characterized in classification calculated value of G ^ conditional probability _{_{P (F 1 = x k \}} C 1) = m COUnt (F '= X ^ C') + l, _wherein, COMW t (= _¾, Q represents a Class ^C '中中取

^ count(F _i = ^, .) + 1 F _f | The number of times the value indicates the number of values that the feature can take. For the information text according to the Bayesian principle and the naive Bayesian hypothesis, the probability that the information text is classified as G is proportional to a _t = P(C _t ) · Π Ρ ( = ( ) IG ) , where ^d) represents information The value of the feature ^Fi in the text. Therefore, the calculation unit of the BT server obtains the value of =ρ(ς)·ΠΡ( = (^)ΙΦ, and expresses the calculation result as a set?^,^, and the calculation unit can sort the Ρ by its element value, such as descending order. , the ordered set Q is obtained, that is, the larger the value of the element is, the higher the position is. The determining unit can determine that the text belongs to the category corresponding to the element ranked first in the set Q, and can also select the top f The element is used as the probability distribution of its corresponding classification, and its selection rules are pre-configured by the determining unit. For example, CI and C2 respectively represent sports and tourism, and the text d of the web page to be analyzed contains "basketball" and "soccer". Characteristic data such as "travel", in the ordered set Q obtained after the above operation, the element ranked first corresponds to C1, and the element ranked second corresponds to C2, if only the first bit is selected Elements, Then, the determining unit of the BT server determines that the type of the text d of the webpage is C1, that is, the sports class. If the first and second elements are selected, it is determined that the type of the text d of the webpage is both a sports class and a tourism class. . The obtaining unit is further configured to acquire user feature data according to the webpage type of the webpage and the user information;

The obtaining step of the obtaining unit 301 is as follows:

Calculating, according to the click stream data, a word frequency TF, an inverse document frequency IDF of the metadata corresponding to the webpage type, and a product TF-IDF of the TF and the IDF;

The calculation unit of the BT server can calculate the word frequency by the Markov model formula, and the formula is as follows: s

Tf _ai , _P = ^ f(k) P(x _k = ai |x ₀ = p)

k=i

Where ^s is the set of historical effective steps, ai e {a^ .^ aM} is a metadata in the metadata set, and the metadata is data representing key terms of the webpage type, such as the webpage type is sports, The metadata can be "sports", or "soccer", or "basketball", etc., and its metadata is preset by the computing unit of the BT server. The webpage type P accessed by the terminal is the target webpage in the Markov process, and the target webpage is the webpage accessed by the user in a time period, and f(k) is a decaying attenuation factor for expressing the previous step, for example, this embodiment Where f(k) = e-Pkt , the smaller the value of the attenuation factor over time, that is, the smaller the weight, the lower the user's behavior, the lower the reference value of the user behavior. P ( xk = ai|xO = p ) indicates the frequency at which ai appears in a target web page, that is, the word tfai, which indicates the word frequency (TF), and the word frequency indicates how often the entry appears in the document or the target web page. The formula indicates the frequency at which a particular entry appears on a web page that the user has visited from the past to the present. For example, to calculate the word frequency of the word "basketball" in 100 web pages accessed by the user from the past to the present, the BT server According to the above formula, the calculation unit passes the input parameters: al= "basketball", S=100, and counts the size of P (xk = ai|x0 = p). Finally, the term "basketball" can be calculated by the above formula. Word frequency.

After the BT server's calculation unit calculates the word frequency, you also need to bring tfai,p into the formula: Tfidf = t X id: , where _tfi = _tfei , p.

Term Frequency (TF) refers to the frequency at which a given word appears in the file or landing page. In the embodiment of the present invention, the target webpage represents a webpage accessed by the terminal. Inverse Document Frequency (IDF) is a measure of the universal importance of a word. The principle is: If the document or landing page containing a certain entry has fewer IDF, the entry has a good class distinguishing ability. The IDF of a particular word can be obtained by dividing the total number of documents by the logarithm of the quotient obtained from the article containing the word:

Where pi: the total number of files in the database or the total number of landing pages, Ιϋ ^: ί3⁄4 Η: the number of files containing words or the number of landing pages containing words, (ie the number of files _s '≠) if the word is not in the database or target In the web page, the divisor is zero, so it is generally used. Therefore, for the formula ¹ ^^^ = ^{tf x ic} , the high word frequency within a particular file is known, and the word is in the entire file collection. The low file frequency, multiplied to find a high weight TF-IDF. TF-IDF means filtering out common words and retaining important words. The higher the value, the more important the word is.

The calculation unit of the BT server calculates the values of tfai, p and idf by the above formula, and multiplies the two to find the value of tfidf.

The obtaining unit 302 selects one or more TF-IDFs with the largest value calculated by the calculating unit, determines metadata corresponding to the one or more TF-IDF values, and queries the correspondence table between the metadata and the user feature data to obtain User characteristic data corresponding to the metadata.

For example, the acquisition unit of the BT server first acquires all types of web pages accessed by the terminal, and has four specific types of "sports class", "music class", "financial class", and "IT class", and then selects "sports", " Tone Four specific terms, "Le", "Finance", and "IT", respectively determine the value of tfidf, and then from the big to the small 歹 'J, the acquisition unit 302 selects the one with the largest value or the first one. If the order is "sports", "music", "finance", "IT", then the user feature of the user acquired by the acquisition unit 302 is "sports fan", or the user feature is "sports fan" and "user" The specific acquisition process can be obtained by querying the correspondence table between the configuration metadata and the user feature. The user feature can be represented by a character string or a data format commonly used by other computers, and the embodiment is not limited.

The determining unit 302 is configured to query a correspondence table between the user feature data pre-configured by the BT server and the delivery policy data, and determine the delivery policy data corresponding to the user feature data;

It should be noted that the correspondence table is pre-configured by the BT server. For example, a user's user characteristics are "ordinary white-collar/influx/gourmet", and the corresponding delivery strategy is "consumer electronic/meal-escaped characters, integer numbers, and other computer-used data format representations, which are represented by strings. Separately

" white_ collar /fashion— follower/ food- lover" , " electronic- consumption/ group_purchasing_of_food". The cookie of the terminal, so that the content providing server receives the access request sent by the terminal when the next time the webpage is accessed, and then delivers the personalized content to the terminal according to the delivery policy data in the cookie of the terminal.

It should be noted that the webpage accessed by the terminal next time may be the webpage accessed in the previous step, or may be another webpage.

The cookie written by the writing unit 303 is as shown in Table 1.

In the embodiment of the present invention, after the obtaining unit receives the user information sent by the terminal, the acquiring unit performs webpage type analysis on the webpage accessed by the terminal, and obtains the user characteristic data and the corresponding delivery policy according to the user information and the webpage type, by writing Enter the unit's cookie into the terminal for the next content When the server receives the terminal access request, the terminal directly performs personalized service delivery on the terminal by reading the cookie of the terminal. Reduce bandwidth consumption and improve server interaction efficiency.

From the description of the above embodiments, those skilled in the art can clearly understand that the present invention can be implemented in hardware, firmware implementation, or a combination thereof. When implemented in software, the functions described above may be stored in or transmitted as one or more instructions or code on a computer readable medium. Computer readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one location to another. A storage medium may be any available media that can be accessed by a computer. By way of example and not limitation, computer readable media may comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage media or other magnetic storage device, or can be used for carrying or storing in the form of an instruction or data structure. The desired program code and any other medium that can be accessed by the computer. Also. Any connection may suitably be a computer readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable , fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, wireless, and microwaves are included in the fixing of the associated media. As used in the present invention, a disk and a disc include a compact disc (CD), a laser disc, a disc, a digital versatile disc (DVD), a floppy disc, and a Blu-ray disc, wherein the disc is usually magnetically copied, and the disc is The laser is used to optically replicate the data. Combinations of the above should also be included within the scope of the computer readable media.

In summary, the above description is only a preferred embodiment of the technical solution of the present invention, and is not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and scope of the present invention are intended to be included within the scope of the present invention.

Claims

Rights request

1. A method of delivering personalized content, characterized by including:

User behavior analysis: The BT server obtains the user information sent by the terminal during the process of accessing the web page, analyzes the web pages accessed by the terminal, and obtains the web page type of the web page;

Obtain user characteristic data according to the web page type of the web page and the user information;

Query the correspondence table between the user characteristic data and the delivery strategy data preconfigured by the BT server, and determine the delivery strategy data corresponding to the user characteristic data;

The user characteristic data and the corresponding delivery strategy data are written into the cookie of the terminal, so that after the content providing server receives the access request sent by the terminal the next time it accesses the web page, it can

2. The method according to claim 1, characterized in that: writing the user characteristic data and corresponding delivery strategy data into a cookie of the terminal further includes: setting the domain and creation time of the cookie. and expiration time.

3. The method according to claim 1 or 2, characterized in that, before obtaining the user characteristic data according to the web page type of the web page and the user information, it further includes:

The user information is verified, and the successfully verified user information is reconstructed into click stream data.

4. The method according to claim 3, characterized in that if the verification of the user information fails, the user information is deleted.

5. The method according to claim 3 or 4, characterized in that the obtaining user characteristic data includes:

According to the click stream data, calculate the word frequency TF, inverse document frequency IDF corresponding to the web page type, and the product word frequency - inverse document frequency TF-IDF of the TF and the IDF;

Select one or more TF-IDF values with the largest value, determine the web page type corresponding to the one or more TF-IDF values, query the correspondence table between the web page type and user characteristic data, and obtain the corresponding web page type. User characteristic data.

6. The method according to any one of claims 1 to 5, characterized in that said obtaining the web page type of the web page includes:

Set a collection of types of web pages and a collection of frequencies corresponding to the types of web pages; Obtain the characteristic data of the web page, the characteristic data includes keywords, character spacing and text length corresponding to the type set of the web page;

Calculate the probability of the feature data according to the set of frequencies corresponding to the type of the web page, select the largest one or more probability values among the calculated probability values, and obtain the web page type corresponding to the selected probability value.

7. The method according to any one of claims 1 to 6, the user information includes Hypertext Transfer Protocol HTTP request information, requested web page information and user behavior information.

8. A user behavior analysis BT server, which is characterized by including:

The acquisition unit is used to obtain the user information sent by the terminal during the process of accessing the web page, analyze the web pages accessed by the terminal, and obtain the web page type of the web page;

The acquisition unit is also used to acquire user characteristic data according to the web page type of the web page and the user information;

The determination unit is used to query the correspondence table between the user characteristic data and the delivery strategy data preconfigured by the BT server, and determine the delivery strategy data corresponding to the user characteristic data;

A writing unit configured to write the user characteristic data and the corresponding delivery strategy data into the cookie of the terminal, so that after the content providing server receives the access request sent by the terminal when it next accesses the web page, it can write the user characteristic data and the corresponding delivery strategy data into the cookie according to the terminal The delivery strategy data in the cookie is used to deliver personalized content to the terminal.

9. The BT server according to claim 8, wherein the writing unit writes the user characteristic data and the corresponding delivery strategy data into the cookie of the terminal, and further includes: the writing unit adds The user characteristic data and the corresponding delivery strategy data are sent to the cookie of the terminal, and the domain, creation time and expiration time of the cookie are set.

10. The BT server according to claim 8 or 9, characterized in that the BT server further includes:

A verification unit, used to verify the user information;

The reconstruction unit is used to reconstruct the user information after successful verification into click stream data.

11. The BT server according to claim 10, characterized in that the server further includes:

A calculation unit, configured to calculate the word frequency TF corresponding to the web page type according to the click stream data, Inverse document frequency IDF and the product of the TF and the IDF TF-IDF;

The acquisition unit is also used to select one or more TF-IDF values with the largest value, determine the web page type corresponding to the one or more TF-IDF values, and query the correspondence table between the web page type and user characteristic data. , obtain user characteristic data corresponding to the web page type.

12. The BT server according to any one of claims 8 to 11, characterized in that the BT server further includes:

The setting unit is used to set the type set of web pages and the set of frequencies corresponding to the types of web pages; the obtaining unit is also used to obtain the characteristic data of the web page, and the characteristic data includes the key corresponding to the type set of the web page Words, character spacing and text length;

The acquisition unit is also configured to calculate the probability of the feature data according to the set of frequencies corresponding to the type of the web page, select the largest one or more probability values among the calculated probability values, and obtain the selection The 4 rate values correspond to the web page type.

13. The BT server according to any one of claims 8 to 12, the user information includes Hypertext Transfer Protocol HTTP request information, requested web page information and user behavior information.