CN102591966A - Filtering method of search results in mobile environment - Google Patents

Abstract

The invention discloses a filtering method of search results in a mobile environment. The method comprises the steps of: finely dividing users into different groups according to history position information characteristics of the users; characteristically modeling the users according to the history query records of the users; analyzing history call records of the users, establishing a social intercourse relation network of the users and calculating the social intercourse relation importance among the users; and during search, firstly, filtering the search results based on contents by using an established user characteristic model, secondly, cooperatively filtering the search results with the finely divided user group information and the excavated information of the social intercourse relation network of the users, and thirdly, returning the search results to the users. With the method for excavating the user characteristics and filtering the information, the search results can be better filtered in a personalized way, a mass of unrelated search results can be removed, a result set can be simplified, and the personalized precise search in the mobile environment can be realized.

Description

A method for filtering search results in mobile scenarios

technical field

The invention belongs to the field of information retrieval, and in particular relates to a method for filtering search results in a mobile scene, and the method is suitable for personalized search in a mobile scene.

Background technique

In the past ten years, search engine technology has achieved rapid development, and traditional Internet search has developed quite maturely from technology implementation to business model, and has achieved great success. In recent years, emerging technologies and applications represented by the mobile Internet have emerged continuously, and mobile search is one of the important applications of the mobile Internet.

Due to the limitations of mobile terminal mobility, portability, and screen size, processing power, and available bandwidth, mobile search cannot directly copy the existing Internet search implementation solutions. The main reasons are as follows: (1) Traditional Internet Search engines usually return a large number of results to users, in fact more than half of these results are irrelevant to users in most cases. One of the main reasons is that the search engine simply matches the search keywords without considering other information (such as user context information, personal preferences, etc.), coupled with the surge of information on the Internet, resulting in many "garbage results" As a result, users have to filter by themselves in the search results, which greatly increases the burden on users. In the mobile scenario, due to the limitations of the size of the mobile terminal screen keyboard, processing power, and available bandwidth, the above situation is unacceptable to users. First, a large number of garbage results waste valuable traffic, and second, users flip through search results on mobile terminals. Page screening is very inconvenient, which determines that mobile search must be an accurate search, and as few as possible, accurate results should be returned to the user; (2) For the same search keyword, the traditional Internet search engine can search for all users The returned results are the same. However, different users have different background knowledge, different hobbies, and different information needs. The same keyword may express different meanings to different people, in different fields, at different times and places. This means that what users need is often only a small subset of all search results. The mobility, portability and privacy of mobile terminals enable users to obtain the required information anytime and anywhere, which makes the demand for personalized search more intense. Context (such as time, location, weather and other factors) relevant personalized search.

Therefore, what mobile search needs to achieve is personalized and precise search. At present, domestic mobile search research is still in its infancy, and its implementation technology is still immature compared with the existing Internet search technology. The earlier technology includes vertical search technology, such as mobile phone music search, novel search, etc. Currently, more implementation schemes are used. Combining the existing Internet search technology and related auxiliary technologies, such as information filtering technology, firstly model the characteristics of users, and then use this model to perform personalized filtering on search results, filter out irrelevant results, and realize personalized and accurate search.

Common technologies for user feature modeling include vector space model and ontology model. Vector space model is relatively widely used because of its simple principle and easy implementation.

Commonly used information filtering technologies are content-based filtering technology and collaborative filtering technology. Content-based filtering technology is to extract features from the results, calculate the similarity between the results and the filtering template (user model), and filter according to the set threshold, because it is based on Analysis of the result content can usually achieve a better filtering effect, but the amount of calculation is relatively large. Collaborative filtering technology is based on the idea that people of the same type usually have the same interest preferences, and collaborative filtering of user search results is performed through users with similar interests to the current user. This technology has achieved good development in the field of e-commerce. and apply.

Contents of the invention

The purpose of the present invention is to provide a method for filtering search results in a mobile scene. The method establishes a user characteristic model and a user social network by mining user data (user historical location information, historical call records, etc.), and based on the user characteristic model and user Social networks perform content-based filtering and collaborative filtering on search results to filter out irrelevant search results and realize personalized and precise search in mobile scenarios, which is valuable for improving mobile search user experience and user stickiness.

A method for filtering search results in a mobile scene provided by the present invention, the method includes the following steps:

Step 1 For the initial result sets R ₁ , R ₂ , . . . , R _Z of users U _i , i=1, 2, . The feature vector of R _r is expressed as f _Rr = {(q ₁ , v ₁ ), (q ₂ , v ₂ ),..., (q _d , v _d )}, v _a represents the weight on each dimension; Use the word frequency/inverse document frequency TF/IDF model to calculate f _Rr , the weight v _a on each dimension, for each word q _a in q ₁ , q ₂ ,...q _d , if it does not appear in If R _r is medium, its weight is 0, otherwise it is the TF/IDF value, TF is the number of times it appears in R _r , IDF is the inverse document frequency, and the number z of the results containing the word is counted;

Among them, the IDF value is log(Z/z), Z is the number of initial results to be filtered, the TF/IDF value is the product of TF and IDF, r=1, 2,..., Z, a=1, 2 ,...,d;

The second step is to find similar users of the current user U _i , and select from the following two user sets, one is the group G _g to which the user belongs, and g is the serial number of the group to which the user belongs, and its value ranges from 1 to m, The second is the set of users in the user's social network. Merge these two sets to obtain a set S, record the user in this set as U _is , use the vector cosine angle formula shown in formula I to calculate the user U _i and set S The similarity between each user U _is , as shown in formula II, the smaller the vector angle, the larger the cosine value, the greater the similarity, and vice versa; i represents the serial number of the user, and N represents the number of users , i=1, 2,..., N, f _Ui and f _Uis represent the eigenvectors of U _i and U _is respectively, ψ(U _i , U _is ) represents the degree of relationship between U _i and U _is , if U _is in the social network of U _i , then ψ(U _i , U _is ) takes the corresponding value, otherwise it takes zero value; select the first n users U _i1 , U _i2 ,... , U _i , if less than n, then select all users in S; n is a preset value;

$\mathrm{sim}(u_i,u_{\mathrm{is}})=(1+\mathrm{\ψ}(u_i,u_{\mathrm{is}}))\&Center\; Dot;\cos (f_{u_i},f_{u_{\mathrm{is}}})$ Formula I

$\cos (f_{u_i},f_{u_{\mathrm{is}}})=\frac{f_{u_i}\·f_{u_{\mathrm{is}}}}{\left|\right|f_{u_{\mathrm{is}}}\left|\right|\·\left|\right|f_{u_{\mathrm{is}}}\left|\right|}$ Formula II

Step 3 Filter based on content:

For each initial result R _r to be filtered, the similarity between it and the user U _i is calculated sequentially using formula III, f _Ui and f _Rr represent the feature vectors of U _i and R _r respectively; according to the similarity according to the preset Threshold ζ filtering, filtering out the initial results whose similarity is smaller than the threshold ζ to obtain an intermediate result set R _r , r=1, 2, ..., Z _ζ , and the intermediate results obtained by filtering are arranged in the original sequence;

$\mathrm{sim}(u_i,R_r)=\cos (f_{u_i},f_{R_r})$ Formula III

in, $\cos (f_{u_i},f_{R_r})=\frac{f_{u_i}\&Center\; Dot;f_{R_r}}{\left|\right|f_{u_i}\left|\right|\&Center\; Dot;\left|\right|f_{R_r}\left|\right|}$

和

分别代表U_is与U_i，U_is与R_r之间的相似度；The second step is to perform collaborative filtering on the intermediate result set R _r , r=1, 2, ..., Z _ζ , using the n most similar users U _i1 , U _i2 , ..., U _{i η} of the user U _i , to The intermediate result R _r ,, calculate the similarity sim'(U _i , R _r ) according to formula IV for collaborative filtering, where,

and

represent the similarity between U _is and U _i , U _is and R _r respectively;

${\mathrm{sim}}^{\′}(u_i,R_r)=\underset{\mathrm{the\; s}=1}{\overset{\mathrm{\η}}{\mathrm{\Σ}}}(\cos (f_{u_{\mathrm{is}}},f_{u_i})\&Center\; Dot;\cos (f_{u_{\mathrm{is}}},f_{R_r}))$ Formula IV

Rank _r ＝θ·r+(1-θ)·sim′(U _i , R _r ) Formula V

According to sim′(U _i , R _r ), perform collaborative filtering according to the preset threshold ε, filter out the intermediate results whose similarity is less than ε, and obtain the temporary result set R _r , r=1, 2, ..., Z _ε , r represents its order in the temporary result set, which is 1, 2, ..., Z _ε , for temporary R _r , use the formula V to calculate its order r and sim with the preset weighting coefficient θ The weighted sum of ′(U _i , R _r ) is used as the final result to rank Rank _r , and the temporary result set R _r is reordered based on this ranking to obtain the final result, which is returned to the user, and the filtering process ends.

The search result filtering method under the mobile scene provided by the present invention comprehensively adopts data mining methods (classification, clustering), and is based on a content filtering algorithm and a collaborative filtering algorithm. Specifically, the present invention has the following effects and advantages:

(1) High accuracy. The present invention innovatively analyzes user social network information, and simultaneously performs collaborative filtering on the basis of traditional content-based filtering, which greatly improves accuracy.

(2) Strong adaptability, the present invention considers the diversity of mobile user groups and individuals, and can well adapt to the personalized needs of various user groups and individuals.

(3) high scalability, the filtering method provided by the present invention can not only be used for mobile search, but also can be used for its mobile Internet application, precise advertisement placement, etc., and the user characteristic modeling method can also be applied to customer relationship management (CRM) wait.

Description of drawings

Fig. 1 is the overall flowchart of the inventive method;

Figure 2 is a simplified diagram of the frequency of changes in the historical location of mobile users;

Fig. 3 is the flowchart of clustering by location of mobile users;

Fig. 4 is a structural diagram of a mobile user social network;

Fig. 5 is a detailed filtering flow chart of mobile search results.

Detailed ways

The present invention will be described in detail below in conjunction with the accompanying drawings.

A method for filtering search results under a mobile scene provided by the present invention, as shown in FIG. 1 , is first a filtering preprocessing stage, mainly including user subdivision, constructing a user characteristic model and constructing a user social network, respectively corresponding to the following steps (1 ) to step (3), and then the result filtering stage, corresponding to the following step (4). The specific processing steps are as follows:

1. Filtration pretreatment stage, comprising the following steps (1) to (3).

(1) User segmentation, using data mining methods to segment users, the user data sets provided by existing telecom operators, which collect a large amount of user data, such as user historical location information, historical call records, user Historical query records and browsing records, historical business data, etc., the present invention mainly uses the user's historical location information to subdivide the user, and the specific steps are as follows:

(a) The users are divided according to the user’s historical position change frequency. The user’s historical position information records the user’s historical position L and corresponding time information T. The position information L is recorded in the data set in the form of latitude and longitude, such as (30.2332, 114.3243 ), the time information T is recorded in the form of time points, and the longitude and latitude of the two adjacent historical positions of the user are known, and the distance can be easily calculated by using the longitude and latitude distance formula (formula (1)), and the longitude and latitude of the first position L ₁ is (lon ₁ , lat ₁ ), the latitude and longitude of the second location L ₂ is (lon ₂ , lat ₂ ), according to the benchmark of the 0-degree meridian, the east longitude takes a positive value, the west longitude takes a negative value, and the north latitude is (90°- lat) into the calculation, the south latitude is calculated according to (90°+lat), and the distance between two points can be calculated by formula (1).

C＝sin(lat ₁ )·sin(lat ₂ )·cos(lon ₁ -lon ₂ )+cos(lat ₁ )·cos(lat ₂ )

$\mathrm{<span\; class="notranslate">\; dis</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; R</span>}<span\; class="notranslate">\; \&Center\; Dot;</span>\mathrm{<span\; class="notranslate">\; arccos</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&Center\; Dot;</span>\frac{\mathrm{<span\; class="notranslate">\; \&pi;</span>}}{\mathrm{<span\; class="notranslate">\; 180</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

For each user U _i , (i=1, 2, ..., N), calculate the accumulative change frequency F _i , (i=1, 2,. .., N), where N represents the number of users.

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&Delta;T</span>}}\underset{\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; m</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; |</span>\frac{\mathrm{<span\; class="notranslate">\; dis</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; L</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

As shown in formula (2), (L ₁ , T ₁ ), (L ₂ , T ₂ ), ..., (L _M , T _M ) are users U _i , (i=1, 2, ... , N) The historical location information within the most recent period of ΔT, (L _k-1 , T _k-1 ) and (L _k , T _k ) are two adjacent historical location and time information of the user, Dis(L _k , L _k-1 ) and T _k −T _k-1 are respectively the difference between the historical location distance and time of two adjacent times. M represents the number of historical positions of the current user, and k represents the sequence number of the historical positions.

Count the F of all users to obtain the overall range range Ω of F, and divide Ω into several sub-intervals Ω ₁ , Ω ₂ , ..., Ω _n , where n represents the number of user groups, and these sub-intervals use F to represent different users Groups, users are divided into corresponding sub-intervals according to their F, as shown in Figure 2, user A has a higher F, and may be a business person who travels frequently. If user B's F is low, he may be in a fixed location for a long time, for example, he may be a student in a certain college. In this way, according to the location change frequency F, a preliminary division of users is carried out, and users are divided into different groups. The groups Ω ₁ , Ω ₂ , ..., Ω _n . Dividing Ω can be divided equally, or a division standard can be preset by the system.

(b) Next, for each Ω _j , (j=1, 2, ..., n, j represents the serial number of the group), the users in each group are clustered according to the historical location information, and the users with adjacent positions are clustered into one group , related investigations and studies have shown that geographically adjacent users have similar user characteristics to a certain extent, and the k-means clustering algorithm is used for each user in Ω _j , (j=1, 2,..., n) Clustering, the steps are as follows:

(b1) First calculate the central position O _{i of each user U i} , (i=1 _{, 2, ..., N) in the historical position within ΔT time, and cluster the users according to O i ;} i represents the user serial number;

(b2) Randomly select k users from Ω _j , (j=1, 2, ..., n), each user U _q , (q = 1, 2, ..., k) represents an initial User cluster C _q , (q=1, 2, ... k), whose O _q , (q = 1, 2, ..., k) represents the initial center of the user cluster;

(b3) For each remaining user in Ω _j , (j=1, 2, ..., n), calculate its center with each user cluster C _q , (q = 1, 2, ... k) O _q , the distance (longitude and latitude distance formula) of (q=1, 2, ..., k), assign it to the nearest user cluster;

(b4) Then recalculate the new central value O _q of each user cluster, (q=1, 2, . . . , k), and replace the old central value. Calculate the value of criterion function E _j according to formula (3), if the value of E _j converges, the clustering process ends, otherwise, go to step b3.

${\mathrm{E.}}_j=\underset{q=1}{\overset{k}{\mathrm{\&Sigma;}}}\underset{u\&Element;{\mathrm{\&Omega;}}_j}{\mathrm{\&Sigma;}}\mathrm{dis}(u,C_q),$ (j=1, 2, ... n) (3)

As shown in formula (3), Dis(U, C _q ) represents the user and user cluster C _q in Ω _j , (j=1, 2, ..., n), (q = 1, 2, .. .k) distance from the center O _q , (q=1, 2, . . . , k).

Clustering to obtain compact user clusters, so that based on the division of Ω ₁ , Ω ₂ , ..., Ω _n , the users are further divided into smaller groups G ₁ , G ₂ , ..., G _m , Implement user segmentation.

(2) Build a user characteristic model, the user's historical query records well characterize the user's interest characteristics, by analyzing the user's historical query records, adopt the vector air-entry model to carry out feature modeling to the user, and its steps include:

(a) Count all historical query records of all users within ΔT time, and obtain d different words q ₁ , q ₂ , ..., q _d as the d dimensions of the vector space, and the user's feature vector is expressed as f _Ui = {(q ₁ , v ₁ ), (q ₂ , v ₂ ), ..., (q _d , v _d )}, (i=1, 2, ..., N), v _a , (a=1, 2, . . . , d) represents the weight of each dimension.

(b) Using TF/IDF (term frequency/inverse document frequency) model, for each user U _i , (i=1, 2, . . . , N), calculate the weight of each dimension of its feature vector. For each word q _a in q ₁ , q ₂ , ..., q _d , (a=1, 2, ..., d), if it does not appear in the user's historical query records, then its corresponding The weight v _a , (a=1, 2, ..., d) is 0, otherwise it is the TF/IDF value, TF is the word frequency, here is the number of times the word appears in the user's historical query records, and IDF is the inverse Document frequency, counting the number D of users who have the word in the historical query records, the IDF value is log(N/D), N is the number of all users, and the TF/IDF value is the product of TF and IDF.

(3) Mining the user's social network information, analyzing the user's historical call records, for each user U _i , (i=1, 2, ..., N), its social network is presented as a star centered on the user Topological diagram, as shown in Figure 3, the central node B represents the user himself, the star nodes A, C, D, E, F, G, etc. represent the users who have call records with B, and the weight ψ of the edge represents the degree of relationship between users , this step is mainly to estimate the value of ψ.

The user's historical call record data records the call records between all users, including the id numbers of both parties), call start time, call end time, etc., for each user U _i , (i=1, 2,... , N), analyze its call record within ΔT time, for each user u _x who has a call record with it, (x=1, 2, ..., e, e represents the number of users who have a call record with it), By analyzing the total number of calls α, total call duration β, and call rules γ between it and U _i , (i=1, 2, ..., N) within ΔT, and comprehensively analyzing these factors, it can be roughly inferred that U _i , ( The relationship degree ψ _ix between i=1, 2, . . . , N) and u _x , (x=1, 2, . . . , e).

The total number of calls α and the total call duration β are relatively easy to obtain, but they are overall statistics, relatively simple, and can only roughly estimate the relationship between users in general, while ignoring important details, such as Whether the distribution of each call event over time is uniform, whether it is uniform overall or locally, etc., so the characteristic factor of call rule γ is also introduced here to represent U _i , (i=1, 2, ..., N) and u _x , the degree of relationship between (x=1, 2, ..., e), by statistically analyzing the time distribution characteristics of all call events within the time ΔT, borrowing the idea of variance, such as formula (4) (5) (6), t _h , (h=1, 2, ..., α) is the start time of each call, Δt _h is the time difference between two adjacent call records, S _t is its variance, and γ is inversely proportional to S _t , as shown in formula (6), a small variance means that the calls within this period are relatively regular, and γ is relatively large, and vice versa.

Δt _h =t _h -t _h-1 , (h=2, 3, . . . , α) (4)

$\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; \&Delta;t</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\underset{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&Delta;t</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 5</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\underset{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 2</span>}}{\overset{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; \&Delta;t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&Delta;t</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 6</span>}<span\; class="notranslate">\; )</span>$

$\mathrm{<span\; class="notranslate">\; \&gamma;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 7</span>}<span\; class="notranslate">\; )</span>$

Normalize the calculated α, β, γ to obtain a value between 0 and 1, ψ _ix , (i=1, 2, ..., N, x = 1, 2, ... , the value of e) is calculated by formula (8), which is a weighted value obtained by comprehensively considering α, β, γ. In formula (8), 0≤λ ₁ ≤1, 0≤λ ₂ ≤1, 0≤ λ ₃ ≤ 1, and λ ₁ +λ ₂ +λ ₃ =1, and its default value is 1/3 of the mean value.

ψ _ix =λ ₁ ·α+λ ₂ ·β+λ ₃ ·γ, (λ ₁ +λ ₂ +λ ₃ =1) (8)

In this way, through the analysis and calculation of this step, the social network information of each user U _i , (i=1, ₂ , . 1, 2, ..., e) the degree of relationship between ψ _ix .

(4) Search result filtering. Steps (1) to (3) above are all preparatory stages for the search result filtering service of this step. The user characteristic model established in step (2) is used to perform search results based on Content filtering, user segmentation in step (1) and user social network information mined in step (3) are used for collaborative filtering of search results.

In this step, content-based filtering is first performed on search results, and then collaborative filtering is performed. To personalize and refine search results.

User U _i , (i=1, 2, ..., N) submits a search Q, the search request is first processed by the existing Internet search engine, and the existing Internet search engine returns an initial result set to the search Q, the result The set is usually large, select the first φ results in the result set to filter, if there are less than φ, select all the initial result sets as the result sets to be filtered R ₁ , R ₂ ,..., R _Z , φ is an empirical value, preset by the system, for example, it is set to 300, and Z is the number of results to be filtered. The result filtering process is shown in Figure 5, and the steps are as follows:

(a) To filter the result sets R ₁ , R ₂ , ..., R _Z , establish feature vectors, use the d-dimensional vector space established in step (2) to establish feature vectors for these results, R _r (r=1, 2, ..., Z) is expressed as f _Rr = {q ₁ , v ₁ ), (q ₂ , v ₂ ), ..., (q _d , v _d )}, (r=1, 2, . . . , Z), v _a , (a=1, 2, . . . , d) represent weights on each dimension. Also use the TF/IDF (term frequency/inverse document frequency) model used in step (2) to calculate f _Rr , the weight v _a of (r=1, 2, ..., Z) on each dimension, (a=1,2,...,d), for each word q _a in q ₁ , q ₂ ,...q _d , (a=1,2,...,d), if its If it does not appear in R _r , (r=1, 2, ..., Z), its weight is 0, otherwise it is its TF/IDF value, and TF is its value in R _r , (r=1, 2, ..., Z), IDF is the inverse document frequency, and counts the number of results containing the word z, the IDF value is log(Z/z), Z is the number of all results, and the TF/IDF value is TF Multiplied by IDF.

(b) Next, look for similar users of the current user U _i , (i=1, 2, ..., N), and select from two user sets, one is the group G _g to which the user belongs in step (1), g is the serial number of the group to which the user belongs, and its value range is from 1 to m. The second is the set of users in the user social network established in step (3), and these two sets are merged (there may be repeated users ) to get a set S, and select several similar users from the set S.

$\mathrm{<span\; class="notranslate">\; sim</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; is</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&psi;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; is</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; is</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; )</span>$

$\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; is</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; \&CenterDot;</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; is</span>}}}}{<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; is</span>}}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; \&CenterDot;</span><span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; is</span>}}}<span\; class="notranslate">\; |</span><span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$

In formula (10), || || represents the modulus of the vector.

(5) Calculate the similarity between U _i , (i=1, 2, ..., N) and each user U _is in the set S by using the vector cosine angle formula shown in formula (10), as As shown in formula (9), the smaller the angle between the vectors, the larger the cosine value, and the larger the similarity, and vice versa. f _Ui and f _Uis represent the feature vectors of U _i and U _is respectively, ψ(U _i , U _is ) represents the degree of relationship between U _i and U _is , if U _is in the social network of U _i , then ψ( U _i , U _is ) take corresponding values, otherwise take zero value. Select the _first n users U _i1 , U _i2 , . η is an empirical value, preset by the system, such as its default value can be 10.

(c) Then start to filter the results. The filtering process is divided into two stages, the content-based filtering stage and the collaborative filtering stage:

(c1) First, based on content filtering, for each initial result R _r to be filtered in (a), (r=1, 2, ..., Z), sequentially calculate its relationship with user U _i , (i=1, 2, ..., N), similarly, use formula (10) to calculate the similarity between the two, as shown in formula (11), f _Ui and f _Rr represent U _i and R _r respectively eigenvectors of . Filter according to the threshold ζ according to the similarity, filter out the results with a similarity smaller than ζ, and obtain the intermediate result set R _r , (r=1, 2, ..., Z _ζ ), the intermediate results obtained by filtering are in the original order arrangement. Threshold ζ is an empirical value, preset by the system, 0≤ζ≤1, and its default value can be set to 0.65.

$\mathrm{<span\; class="notranslate">\; sim</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; )</span>$

和

分别代表U_is与U_i，U_is与R_r之间的相似度。(c2) Next, perform collaborative filtering on the intermediate result set R _r , (r=1, 2, ..., Z _ζ ). Collaborative filtering is based on the idea that similar users usually have similar interests. _{The user comes to carry out collaborative recommendation for the current user, using the n most similar users U i1 ,} U _i2 , . . . , U _{i η} , for the intermediate results R _r , (r=1, 2, ..., Z _ζ ), calculate the similarity sim′(U _i , R _r ) according to formula (12) to perform collaborative filtering, where, using Equation (10) vector cosine angle formula,

and

represent the similarity between U _is and U _i , U _is and R _r respectively.

${\mathrm{<span\; class="notranslate">\; sim</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; \&eta;</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; is</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&CenterDot;</span>\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; u</span>}}_{\mathrm{<span\; class="notranslate">\; is</span>}}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; f</span>}}_{{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}}<span\; class="notranslate">\; )</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">\; 12</span>}<span\; class="notranslate">\; )</span>$

Rank _r = θ·r+(1-θ)·sim'(U _i , R _r ) (13)

According to sim′(U _i , R _r ), perform collaborative filtering according to the threshold ε, filter out the results with similarity less than ε, and obtain a temporary result set R _r , (r=1, 2, ..., Z _ε ), r Represents its order in the temporary result set, which is 1, 2, ..., Z _ε ), for R _r , (r=1, 2, ..., Z _ε ), the weighting coefficient θ is used to calculate its The weighted sum of sequence r and sim'(U _i , R _r ) is used as the final result to rank Rank _r , as shown in formula (13), and the ranking pair R _r , (r=1, 2, ..., Z _ε ) re-sort, get the final result, return to the user, and the filtering process ends. Both the threshold ε and the weighting coefficient θ are empirical values, preset by the system, 0≤ε≤1, 0≤θ≤1, the default value of ε can be set to 0.85, and the default value of θ can be set to 0.5.

The present invention is not limited to the above-mentioned specific embodiments, and those skilled in the art can implement the present invention by using other various specific embodiments according to the disclosed content of the present invention. Changes or modified designs all fall within the protection scope of the present invention.

Claims (7)

1. A method for filtering search results under a mobile scene, the method comprising the steps of: Step 1 For the initial result sets R ₁ , R ₂ , . . . , R _Z of users U _i , i=1, 2, . The eigenvector of R _r is expressed as f _Rr = {q ₁ , v ₁ ), (q ₂ , v ₂ ),..., (q _d , v _d )}, v _a represents the weight on each dimension; use The term frequency/inverse document frequency TF/IDF model calculates f _Rr , the weight v _a on each dimension, for each word q _a in q ₁ , q ₂ ,...q _d , if it does not appear in R _r , medium, its weight is 0, otherwise its TF/IDF value, TF is the number of times it appears in R _r , IDF is the inverse document frequency, count the number z of the results containing the word; Among them, the IDF value is log(Z/z), Z is the number of initial results to be filtered, the TF/IDF value is the product of TF and IDF, r=1, 2,..., Z, a=1, 2 ,...,d; The second step is to find similar users of the current user U _i , and select from the following two user sets, one is the group G _g to which the user belongs, and g is the serial number of the group to which the user belongs, and its value ranges from 1 to m, The second is the set of users in the user's social network. Merge these two sets to obtain a set S, record the user in this set as U _is , use the vector cosine angle formula shown in formula I to calculate the user U _i and set S The similarity between each user U _is , as shown in formula II, the smaller the vector angle, the larger the cosine value, the greater the similarity, and vice versa; i represents the serial number of the user, and N represents the number of users , i=1, 2,..., N, f _Ui and f _Uis represent the eigenvectors of U _i and U _is respectively, ψ(U _i , U _is ) represents the degree of relationship between U _i and U _is , if U _is in the social network of U _i , then ψ(U _i , U _is ) takes the corresponding value, otherwise it takes zero value; select the first n users U _i1 , U _i2 ,... , U _i , if less than n, then select all users in S; n is a preset value; $\mathrm{sim}(u_i,u_{\mathrm{is}})=(1+\mathrm{\&psi;}(u_i,u_{\mathrm{is}}))\&Center\; Dot;\cos (f_{u_i},f_{u_{\mathrm{is}}})$ Formula I $\cos (f_{u_i},f_{u_{\mathrm{is}}})=\frac{f_{u_i}\&Center\; Dot;f_{u_{\mathrm{is}}}}{\left|\right|f_{u_{\mathrm{is}}}\left|\right|\&CenterDot;\left|\right|f_{u_{\mathrm{is}}}\left|\right|}$ Formula II Step 3 Filter based on content: For each initial result R _r to be filtered, the similarity between it and the user U _i is calculated sequentially using formula III, f _Ui and f _Rr represent the feature vectors of U _i and R _r respectively; according to the similarity according to the preset Threshold ζ filtering, filtering out the initial results whose similarity is smaller than the threshold ζ to obtain an intermediate result set R _r , r=1, 2, ..., Z _ζ , and the intermediate results obtained by filtering are arranged in the original sequence; $\mathrm{sim}(u_i,R_r)=\cos (f_{u_i},f_{R_r})$ Formula III in, $\cos (f_{u_i},f_{R_r})=\frac{f_{u_i}\&Center\; Dot;f_{R_r}}{\left|\right|f_{u_i}\left|\right|\&CenterDot;\left|\right|f_{R_r}\left|\right|}$ The second step is to perform collaborative filtering on the intermediate result set R _r , r=1, 2, ..., Z _ζ , using the n most similar users U _i1 , U _i2 , ..., U _{i η} of the user U _i , to The intermediate result R _r ,, calculate the similarity sim'(U _i , R _r ) according to formula IV for collaborative filtering, where,

and

represent the similarity between U _is and U _i , U _is and R _r respectively; ${\mathrm{sim}}^{\&prime;}(u_i,R_r)=\underset{\mathrm{the\; s}=1}{\overset{\mathrm{\&eta;}}{\mathrm{\&Sigma;}}}(\cos (f_{u_{\mathrm{is}}},f_{u_i})\&CenterDot;\cos (f_{u_{\mathrm{is}}},f_{R_r}))$ Formula IV Rank _r ＝θ·r+(1-θ)·sim′(U _i , R _r ) Formula V According to sim′(U _i , R _r ), perform collaborative filtering according to the preset threshold ε, filter out the intermediate results whose similarity is less than ε, and obtain the temporary result set R _r , r=1, 2, ..., Z _ε , r represents its order in the temporary result set, which is 1, 2, ..., Z _ε , for temporary R _r , use the formula V to calculate its order r and sim with the preset weighting coefficient θ The weighted sum of ′(U _i , R _r ) is used as the final result to rank Rank _r , and the temporary result set R _r is reordered based on this ranking to obtain the final result, which is returned to the user, and the filtering process ends. 2. the search result filtering method under the mobile scene according to claim 1, is characterized in that: the initial result set in the 1st step obtains in the following manner: For user U _i to submit a search Q, the search request is first processed by the existing Internet search engine, and the existing Internet search engine returns an initial result set to the search Q, and selects the first φ results in the result set to filter, if If there are less than φ items, select all the initial result sets as the result sets to be filtered R ₁ , R ₂ ,..., R _Z , φ is preset by the system, and Z is the number of results to be filtered. 3. the search result filtering method under the mobile scene according to claim 1, is characterized in that: the 1st step obtains the feature vector of result to be filtered according to the following manner: Count all historical query records of all users within ΔT time, and obtain d different words q ₁ , q ₂ , ..., q _d as the d dimensions of the vector space, and the user's feature vector is expressed as f _Ui = {q ₁ , v ₁ ), (q ₂ , v ₂ ), ..., (q _d , v _d )}, i=1, 2, ..., N, v _a , a=1, 2, ..., d represents the weight of each dimension. 4. The search result filtering method under the mobile scene according to claim 1, characterized in that: in the second step, the most similar user is obtained in the following manner: Step 4.1 Find similar users of the current user U _i , and merge the group G _g to which the user belongs with the set of users in the user's social network to obtain a set S, where g is the serial number of the group to which the user belongs, and its value range is 1 to m, m represents the number of groups; Step 4.2 Use formula VI to calculate the similarity sim(U _i , U _is ) between U _i and each user U _is in the set S, f _Ui and f _Uis represent the feature vectors of U _i and U _is respectively, ψ (U _i , U _is ) represents the degree of relationship between U _i and U _is , if U _is in U _i 's social network, then ψ(U _i , U _is ) takes the corresponding value, otherwise it takes zero value; press Select the first n users U _i1 , U _i2 , _{. .} . $\mathrm{sim}(u_i,u_{\mathrm{is}})=(1+\mathrm{\&psi;}(u_i,u_{\mathrm{is}}))\&Center\; Dot;\cos (f_{u_i},f_{u_{\mathrm{is}}})$ Formula VI in, $\cos (f_{u_i},f_{u_{\mathrm{is}}})=\frac{f_{u_i}\&CenterDot;f_{u_{\mathrm{is}}}}{\left|\right|f_{u_{\mathrm{is}}}\left|\right|\&Center\; Dot;\left|\right|f_{u_{\mathrm{is}}}\left|\right|}.$ 5. the method for filtering search results under the mobile scene according to claim 4, characterized in that: in the 4.1 step, the group _G to which the user belongs is obtained in the following manner: Step 5.1 Divide the users according to the user's historical location change frequency. The user's historical location information records the user's historical location information L and corresponding time information T. The historical location information L is recorded in the data set in the form of latitude and longitude, and the time information T Recorded in the form of time points, the longitude and latitude of the two adjacent historical locations of the user are known, and the distance is calculated using the longitude and latitude distance formula; For each user U _i , calculate the accumulative change frequency F _ij of its historical position within the latest period of time ΔT according to formula VII: $f_i=\frac{1}{\mathrm{\&Delta;T}}\underset{1}{\overset{m}{\mathrm{\&Sigma;}}}\left|\frac{\mathrm{dis}(L_k,L_{k-1})}{T_k-T_{k-1}}\right|$ VII (L ₁ , T ₁ ), (L ₂ , T ₂ ), ..., (L _M , T _M ) are the historical location information of user U _i within the latest period ΔT, (L _k-1 , T _{k -1} ) and (L _k , T _k ) are the user's two adjacent historical location and time information, and Dis(L _k , L _k-1 ) and T _k -T _k-1 are the two adjacent histories respectively The difference between location distance and time; M represents the number of historical locations of the current user, and k represents the sequence number of historical locations; Step 5.2 counts the accumulative change frequency F of all users' historical positions, and obtains the overall range interval Ω of F, and divides Ω into several sub-intervals Ω ₁ , Ω ₂ , ..., Ω _n , where n represents the number of user groups, these Sub-intervals use F to represent different user groups, users are divided into corresponding sub-intervals according to their F, and users are divided into different groups Ω ₁ , Ω ₂ , ..., Ω _n ; Step 5.3 Cluster the users in each Ω _j according to the historical location information, cluster the users with adjacent locations into one group, and then further divide the users into smaller groups G ₁ , G ₂ ,..., G _m , j=1, 2, . . . , n, j represents the serial number of the group. 6. according to the search result filtering method under the mobile scene described in claim 5, it is characterized in that: the 5.3rd step adopts k-means clustering algorithm to carry out clustering to the user in each Ω _{j li} , and step is as follows: (b1) First calculate the center position O _i of the historical position of each user U _i in the most recent period of ΔT, and cluster the users according to the center position O _i ; i represents the serial number of the user; (b2) Randomly select k users from Ω _j , each user U _q represents an initial user cluster C _q , and its center position O _q represents the initial center of the user cluster, q 1, 2, ..., k ; (b3) For each remaining user in Ω _j , calculate the distance between it and the central position O _q of each user cluster C _q , and assign it to the nearest user cluster; (b4) Then recalculate the new center position O _q of each user cluster, and replace the old center value; calculate the value of the criterion function E _j according to formula VIII, if the value of E _j converges, the clustering process ends, otherwise, Go to step b3; ${\mathrm{E.}}_j=\underset{q=1}{\overset{k}{\mathrm{\&Sigma;}}}\underset{u\&Element;{\mathrm{\&Omega;}}_j}{\mathrm{\&Sigma;}}\mathrm{dis}(u,C_q),j=\mathrm{1,2},...\mathrm{no}$ Formula VIII In Formula VIII, Dis(U, C _q ) represents the distance between the user in Ω _j and the user cluster C _q , and the center position O _q ; (b5) Clustering to obtain compact user clusters, so that based on the division of Ω ₁ , Ω ₂ , ..., Ω _n , the users are further divided into smaller groups G ₁ , G ₂ , ..., G _m , realize user segmentation. 7. The search result filtering method under the mobile scene according to claim 4, characterized in that: in the 4.1 step, the user social network is constructed in the following manner: Step 7.1 uses the word frequency/inverse document frequency TF/IDF model to calculate the weight of each dimension of its feature vector for each user U _i ; for each word q in q ₁ , q ₂ ,...,q _{d a,} , if it does not appear in the user's historical query records, its corresponding weight v _a is 0, otherwise it is its TF/IDF value, TF is the word frequency, IDF is the inverse document frequency, count those that appear in the historical query records The number D of users who have passed the word, the IDF value is log(N/D), N is the number of all users, and the TF/IDF value is the product of TF and IDF; Step 7.2 Analyze the call records of each user U _i within the latest period of ΔT, and analyze the total number of calls α and total call duration β of each user u _x with U _i within ΔT. , call law γ , use formula IX to calculate the degree of relationship between U _i and u _x ψ _ix ; ψ _ix =λ ₁ ·α+λ ₂ ·β+λ ₃ ·γ Formula IX In the formula, 0≤λ ₁ ≤1, 0≤λ ₂ ≤1, 0≤λ ₃ ≤1, and λ ₁ +λ ₂ +λ ₃ =1 $\mathrm{<span\; class="notranslate">\; \&gamma;</span>}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}}$ ${\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; t</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\underset{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 2</span>}}{\overset{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; \&Delta;t</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&Delta;t</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}$ Δt _h =t _h -t _h-1 , h=2, 3, . . . , α $\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; \&Delta;t</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}\underset{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; \&alpha;</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; \&Delta;t</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; .</span>$

Images