CN103716471A - User call behavior model generating method applicable to spam voice filtering - Google Patents

Abstract

The invention discloses a user call behavior model generating method applicable to spam voice filtering. The user call behavior model generating method comprises the following steps that: call interactive behavior characteristics (CI) are established and are used for describing related behavior characteristics of specified users which are adopted as a calling user and a called user, wherein the call interactive behavior characteristics (CI) further include incoming call/outgoing call ratios, call interactive recording characteristic value and interaction strength, and distribution of the incoming call/outgoing call ratios and the call interactive recording characteristic value and interaction strength; call frequency and distribution thereof FCD are established and are used for describing the characteristics of call time in call records, wherein the characteristics of the call time in call records include call frequency values and call frequency value time distribution in a specified statistical duration; and call lasting time and distribution thereof DCT are established and are used for describing the characteristics of lasting time in the call records, wherein the characteristics of the lasting time in the call records include call answering refusal proportions, call average lasting time and histogram distribution of the call lasting time. With the user call behavior model generating method applicable to the spam voice filtering of the invention adopted, a technical problem that a camouflaged call behavior of a spam voice transmitter is hard to find can be solved.

Description

Method for generating user call behavior model suitable for filtering junk voice

Technical Field

The invention belongs to the field of junk voice filtering and data mining, and particularly relates to a method for generating a user call behavior model suitable for junk voice filtering.

Background

With the combination of fixed networks, mobile communication networks, and the internet, voice services are widely used. However, due to the influence of the garbage voice, the voice service is hindered by service expansion, and the malicious users need to be limited in time while normal communication of the users is ensured. The existing junk voice filtering technology is mostly improved on the basis of junk mail and junk short message filtering technology, and can play a role in detecting and filtering junk voice to a certain extent. However, the nature of the content of the spam speech is different from that of the spam, and the filtering technology of the spam has certain limitations. The spam is mostly text filtering, and the spam voice content is multimedia information. Spam filtering allows time delays and spam voice has a high real-time requirement.

An effective and reasonable filtering mechanism requires that communication with a calling party and a called party is as less as possible, and a filtering method based on a call model can be adopted. The calling model objectively reflects whether the user is garbage voice or not according to the behavior characteristics of the call made by the user. The existing call model provides a large amount of spam voice characteristics observed from call behaviors, and adopts a method of a decision tree or a Bayesian classifier to realize spam voice filtering. Under the condition that the calling behavior of the user is not changed, the existing calling model is basically mature and effective. However, the existing detection mechanism based on the call model is difficult to find the disguised call behavior of the spammer, and has certain defects.

Disclosure of Invention

In view of the above defects or improvement needs in the prior art, the present invention provides a method for generating a user call behavior model suitable for spam voice filtering, and aims to solve the technical problem in the prior art that it is difficult to find a disguised call behavior of a spammer.

To achieve the above object, according to one aspect of the present invention, a method for generating a user call behavior model suitable for spam voice filtering is provided, which includes the following steps:

(1) establishing a call interaction behavior characteristic CI for describing relevant behavior characteristics of a specified user as a calling user and a called user, wherein the call interaction behavior characteristic CI further comprises three parts of an incoming-outgoing ratio, a call interaction record characteristic value, interaction strength and distribution thereof, and can be specifically expressed as follows:

CI＝{R_in/out，C_out，C_in，C_in/out，F_in/out}

wherein R is_in/outAs a proportion of subscribers as calling and called subscribers, C_outThe number of outgoing calls a subscriber has in all calls with different subscribers only as a calling subscriber, C_inThe number of the user who has the answering action only as the called user in all the calls with different users, C_in/outIs the number of subscribers simultaneously acting as calling and called subscribers in all calls with different subscribers, F_in/outIs the interaction intensity frequency distribution with other users;

(2) establishing call frequency and distribution thereof F_CDThe feature for describing the calling time in the call record includes two parts, namely a calling frequency value in a specified statistical time period and a calling frequency value time distribution, which can be specifically expressed as:

<math> <mrow> <msub> <mi>F</mi> <mi>CD</mi> </msub> <mo>=</mo> <mo>{</mo> <msubsup> <mi>F</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> <mi>T</mi> </msubsup> <msubsup> <mrow> <mo>,</mo> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> </mrow> <mi>out</mi> <mi>T</mi> </msubsup> <mo>}</mo> </mrow> </math>

wherein,

to count the absolute frequency value of calls made by users within a time period,

is the distribution of the frequency of calls of the users within a statistical time period over 12 time slices of 2 hours in length a day.

(3) Set-up call duration and distribution thereof D_CTThe features for describing the duration in the call record include three parts of call rejection ratio, call average duration and histogram distribution of call duration in statistical time period, which can be specifically expressed as:

$D_{\mathrm{CT}}=\{f_E^T,{\mathrm{CT}}_{\mathrm{avg}}^T,{\mathrm{CTD}}^T\}$

wherein,

a probability value for refusal of answering in the call initiated by the user,

average call duration, CTD, for a subscriber^TIs the distribution of call duration.

Preferably, step (1) comprises in particular the following sub-steps:

(1-1) statistical user's historical call interaction behavior characteristic parameter C_out，C_inAnd C_in/outIn order to facilitate the lateral comparison among users, the method further performs normalization processing on the user data:

<math> <mrow> <msub> <mover> <mi>C</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> </msub> <mo>=</mo> <mfrac> <msub> <mi>C</mi> <mi>out</mi> </msub> <mrow> <msub> <mi>C</mi> <mi>out</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mi>in</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> </mrow> </mfrac> </mrow> </math>

<math> <mrow> <msub> <mover> <mi>C</mi> <mo>&OverBar;</mo> </mover> <mi>in</mi> </msub> <mo>=</mo> <mfrac> <msub> <mi>C</mi> <mi>in</mi> </msub> <mrow> <msub> <mi>C</mi> <mi>out</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mi>in</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> </mrow> </mfrac> </mrow> </math>

<math> <mrow> <msub> <mover> <mi>C</mi> <mo>&OverBar;</mo> </mover> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> <mo>=</mo> <mfrac> <msub> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> <mrow> <msub> <mi>C</mi> <mi>out</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mi>in</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> </mrow> </mfrac> </mrow> </math>

wherein,and

the sum of the three is 1.

（1-2) statistics of incoming and outgoing call proportions R_in/outThe calculation formula is as follows:

$R_{\mathrm{in}/\mathrm{out}}=\frac{C_{\mathrm{out}}}{C_{\mathrm{in}}}$

(1-3) counting the call interaction intensity CD of the specified user and other users_in/out. The strength of the call interaction between the user and user j is expressed as

The calculation formula is as follows:

${\mathrm{CD}}_{\mathrm{in}/\mathrm{out}}^j=\mathrm{INT}\left[\log (C_{\mathrm{out}}^j+C_{\mathrm{in}}^j)\right]$

wherein INT [. cndot]The function of rounding is represented by a number of,representing the number of times the user actively calls user j,

indicating the number of times the user answered the call from user j.

(1-4) statistical Call interaction Strength distribution CDD_in/out

CDD_in/out＝{CDN₀，CDN₁，CDN₂，CDN₃}

Among them, CDN_iThe calculation formula of (i ═ 0,1, 2, 3) is as follows:

CDN (content delivery network)_iIs a CD with a value equal to i or greater than i_in/outOf wherein COUNT [ ·]In order to be a function of the count,

indicating the call interaction strength between the user and the user j, and n indicating the number of all contacts of the user.

(1-5) for CDD_in/outNormalization processing is carried out, and the call interaction intensity distribution after normalization is recorded as

The concrete expression is as follows:

<math> <mrow> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> <mo>=</mo> <mo>{</mo> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mn>0</mn> </msub> <mo>,</mo> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mn>1</mn> </msub> <mo>,</mo> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mn>2</mn> </msub> <mo>,</mo> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mn>3</mn> </msub> <mo>}</mo> </mrow> </math>

wherein,

for normalized CDN_i(i ═ 0,1, 2, 3), the calculation formula is:

<math> <mrow> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mi>i</mi> </msub> <mo>=</mo> <mfrac> <msub> <mi>CDN</mi> <mi>i</mi> </msub> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>k</mi> <mo>+</mo> <mn>0</mn> </mrow> <mi>a</mi> </msubsup> <msub> <mi>CDN</mi> <mi>k</mi> </msub> </mrow> </mfrac> <mo>.</mo> </mrow> </math>

preferably, step (2) comprises in particular the following sub-steps:

(2-1) counting the absolute frequency value of the call initiated by the user

$F_{\mathrm{in}/\mathrm{out}}^T=\frac{C_{\mathrm{out}}^T}{T}$

Wherein, T is a designated statistical time period length, and the unit is hour, the time period should start from 0 of the first day and end at 24 of the last day, so T should be an integral multiple of 24 hours;

the number of calls made by the user as the calling user in the time period.

(2-2) counting call frequency distribution parameters of a specified subscriber within a specified time period T

The concrete expression is as follows:

$D_{\mathrm{out}}^T=\{D_{\mathrm{out}}^1,D_{\mathrm{out}}^2,...,D_{\mathrm{out}}^{12},\}$

wherein,

represents the sum of the number of calls in the [2 x (T-1),2 x T) period of each day of the statistical time period T. Each day time can be divided into 12 time slices, each time slice comprises 2 hours, and the 12 time slices specifically comprise: [0,2),[2,4),[4,6),[6,8),[8,10),[10,12),[12,14),[14,16),[16,18),[18,20),[20,22),[22,24).

(2-3) Call frequency distribution parameter to subscribers

Carrying out normalization treatment, wherein the treatment process is as follows:

<math> <mrow> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mi>t</mi> </msubsup> <mo>=</mo> <mfrac> <msubsup> <mi>D</mi> <mi>out</mi> <mi>t</mi> </msubsup> <msubsup> <mi>C</mi> <mi>out</mi> <mi>T</mi> </msubsup> </mfrac> <mo>,</mo> <mrow> <mo>(</mo> <mi>t</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>.</mo> <mo>,</mo> <mn>12</mn> <mo>)</mo> </mrow> </mrow> </math>

the normalized call frequency distribution is then:

<math> <mrow> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mi>T</mi> </msubsup> <mo>=</mo> <mo>{</mo> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mn>1</mn> </msubsup> <mo>,</mo> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mn>2</mn> </msubsup> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>.</mo> <mo>,</mo> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mn>12</mn> </msubsup> <mo>,</mo> <mo>}</mo> <mo>.</mo> </mrow> </math>

(2-4) obtaining the calling frequency with use efficiency and the distribution parameter F thereof_CD

<math> <mrow> <msub> <mi>F</mi> <mi>CD</mi> </msub> <mo>=</mo> <mo>{</mo> <msubsup> <mi>F</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> <mi>T</mi> </msubsup> <mo>,</mo> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mi>T</mi> </msubsup> <mo>}</mo> <mo>.</mo> </mrow> </math>

Preferably, step (3) comprises in particular the following sub-steps:

(3-1) counting the rate of refused answering in the calling initiated by the user

The calculation formula is as follows:

$f_E^T=\frac{{\mathrm{CR}}_{\mathrm{out}}^T}{C_{\mathrm{out}}^T}$

wherein,representing all calls made by the user within a statistical time period,

representing the rejected active call times in the statistical period of time;

(3-2) counting average call duration of the subscriber

The calculation formula is as follows:

<math> <mrow> <msubsup> <mi>CT</mi> <mi>avg</mi> <mi>T</mi> </msubsup> <mo>=</mo> <mfrac> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>r</mi> <mo>=</mo> <mn>1</mn> </mrow> <msubsup> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> <mi>T</mi> </msubsup> </msubsup> <msub> <mi>t</mi> <mi>r</mi> </msub> </mrow> <msubsup> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> <mi>T</mi> </msubsup> </mfrac> </mrow> </math>

wherein,

representing the number of calls in a statistical period of time, t_rRecording the corresponding call duration for the r-th call, wherein the unit is second;

(3-3) histogram distribution of statistical Call duration CTD^TSpecifically, the following are shown:

${\mathrm{CTD}}^T=\{{\mathrm{CT}}_D^0,{\mathrm{CT}}_D^1,...,{\mathrm{CT}}_D^6,{\mathrm{CT}}_D^7\}$

wherein,

the call record proportion of the call duration in a certain time period range is represented, and the specific calculation formula is as follows:

in general, compared with the prior art, the above technical solution contemplated by the present invention can achieve the following beneficial effects:

1. the invention completes the establishment of the user call behavior model suitable for filtering the junk voice and provides a calculation method of the user call behavior characteristics.

2. Aiming at the defect of insufficient original call behavior data, the invention provides a defining mode of combining three enhanced characteristic parameters.

3. The invention is based on the calling interactive characteristics of the calling party and the called party in the calling process, considers the interactive intensity and the distribution condition thereof besides the basic calling-calling ratio, supplements and explains the characteristics of calling interaction, and is more suitable for filtering junk voice.

Drawings

FIG. 1 is a schematic diagram of a method for generating a user call behavior model suitable for spam voice filtering according to the present invention.

FIG. 2 is a detailed flow chart of step (1) of the method of the present invention.

FIG. 3 is a detailed flow chart of step (2) of the method of the present invention.

FIG. 4 is a detailed flow chart of step (3) of the method of the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

The model for describing the user calling behavior provided by the invention comprises a plurality of calling behavior basic parameters, and mainly comprises four parameters, namely a calling user, a called user, calling time, duration and the like. Three different enhanced calling behavior characteristics are further constructed on the basis of the four parameters, wherein the three different enhanced calling behavior characteristics respectively comprise calling interaction characteristics CI, calling frequency and distribution F thereof_CDDuration of calls and distribution thereof D_CTEach call feature is composed of a respective associated call behavior.

As shown in fig. 1, the method for generating a user call behavior model suitable for spam voice filtering according to the present invention includes the following steps:

(1) establishing a call interaction behavior characteristic CI for describing relevant behavior characteristics of a specified user as a calling user and a called user, wherein the call interaction behavior characteristic CI further comprises three parts of an incoming-outgoing ratio, a call interaction record characteristic value, interaction strength and distribution thereof, and can be specifically expressed as follows:

CI＝{R_in/out，C_out，C_in，C_in/out，F_in/out}

wherein R is_in/outFor users as calling and called partiesProportion of households, C_outThe number of outgoing calls a subscriber has in all calls with different subscribers only as a calling subscriber, C_inThe number of the user who has the answering action only as the called user in all the calls with different users, C_in/outIs the number of subscribers simultaneously acting as calling and called subscribers in all calls with different subscribers, F_in/outIs the interaction intensity frequency distribution with other users;

(2) establishing call frequency and distribution thereof F_CDThe feature for describing the calling time in the call record includes two parts, namely a calling frequency value in a specified statistical time period and a calling frequency value time distribution, which can be specifically expressed as:

<math> <mrow> <msub> <mi>F</mi> <mi>CD</mi> </msub> <mo>=</mo> <mo>{</mo> <msubsup> <mi>F</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> <mi>T</mi> </msubsup> <mo>,</mo> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mi>T</mi> </msubsup> <mo>}</mo> </mrow> </math>

wherein,

to count the absolute frequency value of calls made by users within a time period,

is the distribution of the frequency of calls of the users within a statistical time period over 12 time slices of 2 hours in length a day.

(3) Set-up call duration and distribution thereof D_CTFeatures for describing duration in call log including percentage of call rejections, average duration of calls, and histogram score of call duration over statistical time periodThe cloth comprises three parts, which can be specifically expressed as:

$D_{\mathrm{CT}}=\{f_E^T,{\mathrm{CT}}_{\mathrm{avg}}^T,{\mathrm{CTD}}^T\}$

wherein,a probability value for refusal of answering in the call initiated by the user,

average call duration, CTD, for a subscriber^TIs the distribution of call duration.

As shown in fig. 2, step (1) of the method of the present invention comprises the steps of:

201, inquiring all historical call records of the specified user, and counting the total outgoing call times C_outTotal number of incoming calls C_inAnd total number of incoming and outgoing calls C_in/out；

202, C counted in step 201_out，C_inAnd C_in/outAnd (6) carrying out normalization processing. The historical interactive feature values in the normalized call records are recorded as:

and

the calculation formula is as follows:

<math> <mrow> <msub> <mover> <mi>C</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> </msub> <mo>=</mo> <mfrac> <msub> <mi>C</mi> <mi>out</mi> </msub> <mrow> <msub> <mi>C</mi> <mi>out</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mi>in</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> </mrow> </mfrac> </mrow> </math>

<math> <mrow> <msub> <mover> <mi>C</mi> <mo>&OverBar;</mo> </mover> <mi>in</mi> </msub> <mo>=</mo> <mfrac> <msub> <mi>C</mi> <mi>in</mi> </msub> <mrow> <msub> <mi>C</mi> <mi>out</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mi>in</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> </mrow> </mfrac> </mrow> </math>

<math> <mrow> <msub> <mover> <mi>C</mi> <mo>&OverBar;</mo> </mover> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> <mo>=</mo> <mfrac> <msub> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> <mrow> <msub> <mi>C</mi> <mi>out</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mi>in</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> </mrow> </mfrac> </mrow> </math>

wherein,and

the sum of the three is 1;

203, counting the incoming and outgoing call proportion R of the specified user_in/outThe calculation formula is as follows:

$R_{\mathrm{in}/\mathrm{out}}=\frac{C_{\mathrm{out}}}{C_{\mathrm{in}}}$

204, counting the number of outgoing calls and incoming calls between the specified user and other users, and recording as C_outAnd C_in。Representing the number of times the user actively calls user j,

indicating the number of times the user answered the call from user j. Only the statistical results of the outgoing call and incoming call records of the user and other users are saved.

205, according to the result of step 204, further counting the call interaction strength CD of the specified user and other users_in/out。

The call interaction strength between the user and the user j is represented by the following calculation formula:

${\mathrm{CD}}_{\mathrm{in}/\mathrm{out}}^j=\mathrm{INT}\left[\log (C_{\mathrm{out}}^j+C_{\mathrm{in}}^j)\right]$

wherein INT [. cndot]The function of rounding is represented by a number of,representing the number of times the user actively calls user j,

indicating the number of times the user answered the call from user j.

206, statistics of call interaction intensity distribution CDD of specified users_in/out，

CDD_in/out＝{CDN₀，CDN₁，CDN₂，CDN₃}

Among them, CDN_iThe calculation formula of (i ═ 0,1, 2, 3) is as follows:

CDN (content delivery network)_iIs a CD with a value equal to i or greater than i_in/outOf wherein COUNT [ ·]In order to be a function of the count,

indicating the call interaction strength between the user and the user j, and n indicating the number of all contacts of the user.

207, for CDD_in/outNormalization processing is carried out, and the call interaction intensity distribution after normalization is recorded asThe concrete expression is as follows:

<math> <mrow> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> <mo>=</mo> <mo>{</mo> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mn>0</mn> </msub> <mo>,</mo> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mn>1</mn> </msub> <mo>,</mo> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mn>2</mn> </msub> <mo>,</mo> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mn>3</mn> </msub> <mo>}</mo> </mrow> </math>

wherein,

for normalized CDN_i(i ═ 0,1, 2, 3) the formula:

<math> <mrow> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mi>i</mi> </msub> <mo>=</mo> <mfrac> <msub> <mi>CDN</mi> <mi>i</mi> </msub> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>k</mi> <mo>+</mo> <mn>0</mn> </mrow> <mi>a</mi> </msubsup> <msub> <mi>CDN</mi> <mi>k</mi> </msub> </mrow> </mfrac> <mo>.</mo> </mrow> </math>

208, the call interaction feature parameter CI for the specified user is output.

As shown in fig. 3, step (2) of the method of the present invention comprises the steps of:

301, inquiring all historical call records including incoming calls and outgoing calls in a specified time period of a specified user;

302, count absolute frequency values of calls originated by specified users

$F_{\mathrm{in}/\mathrm{out}}^T=\frac{C_{\mathrm{out}}^T}{T}$

Wherein T is the length of a specified statistical time period, the unit is hour, and T is an integral multiple of 24 hours;

the number of calling times of the user as the calling user in the time period;

303, when counting and pointingCall frequency distribution parameter for specified subscriber within time period TThe concrete expression is as follows:

$D_{\mathrm{out}}^T=\{D_{\mathrm{out}}^1,D_{\mathrm{out}}^2,...,D_{\mathrm{out}}^{12},\}$

wherein,

represents the sum of the number of calls in the [2 x (T-1),2 x T) period of each day of the statistical time period T. Each day can be divided into 12 time slices, each time slice comprises 2 hours, and the 12 time slices are as follows: [0,2), [2,4), …, [22, 24);

304, calling frequency distribution parameter to userCarrying out normalization treatment, wherein the treatment process is as follows:

<math> <mrow> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mi>t</mi> </msubsup> <mo>=</mo> <mfrac> <msubsup> <mi>D</mi> <mi>out</mi> <mi>t</mi> </msubsup> <msubsup> <mi>C</mi> <mi>out</mi> <mi>T</mi> </msubsup> </mfrac> <mo>,</mo> <mrow> <mo>(</mo> <mi>t</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>.</mo> <mo>,</mo> <mn>12</mn> <mo>)</mo> </mrow> </mrow> </math>

the normalized call frequency distribution is

<math> <mrow> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mi>T</mi> </msubsup> <mo>=</mo> <mo>{</mo> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mn>1</mn> </msubsup> <mo>,</mo> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mn>2</mn> </msubsup> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>.</mo> <mo>,</mo> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mn>12</mn> </msubsup> <mo>,</mo> <mo>}</mo> <mo>;</mo> </mrow> </math>

305, outputting the calling frequency of the specified user in the specified statistical time and the distribution parameter F thereof_CD。

As shown in fig. 4, step (3) of the method of the present invention comprises the steps of:

401, querying all historical call records, including incoming calls and outgoing calls, of a specified user within a specified time period T;

402, counting the rejected ratio of the user's call

The calculation formula is as follows:

$f_E^T=\frac{{\mathrm{CR}}_{\mathrm{out}}^T}{C_{\mathrm{out}}^T}$

wherein,

representing all calls made by the user during the statistical period,

representing the rejected active call times in the statistical period of time;

403, counting the average call duration of the user

The calculation formula is as follows:

<math> <mrow> <msubsup> <mi>CT</mi> <mi>avg</mi> <mi>T</mi> </msubsup> <mo>=</mo> <mfrac> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>r</mi> <mo>=</mo> <mn>1</mn> </mrow> <msubsup> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> <mi>T</mi> </msubsup> </msubsup> <msub> <mi>t</mi> <mi>r</mi> </msub> </mrow> <msubsup> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> <mi>T</mi> </msubsup> </mfrac> </mrow> </math>

wherein,

representing the number of calls in a given time period T, T_rRecording the corresponding call duration for the r-th call, wherein the unit is second;

404, histogram distribution of statistical Call duration CTD^TSpecifically, the following are shown:

${\mathrm{CTD}}^T=\{{\mathrm{CT}}_D^0,{\mathrm{CT}}_D^1,...,{\mathrm{CT}}_D^6,{\mathrm{CT}}_D^7\}$

wherein,

the call record proportion of the call duration within a certain time period range in the specified time period T is represented, and the specific calculation formula is as follows:

405, outputting the call duration of the specified user in the specified statistical time and the distribution parameter D thereof_CT。

It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (4)

1. A method for generating a user call behavior model suitable for spam voice filtering is characterized by comprising the following steps:

(1) establishing a call interaction behavior characteristic CI for describing relevant behavior characteristics of a specified user as a calling user and a called user, wherein the call interaction behavior characteristic CI further comprises three parts of an incoming-outgoing ratio, a call interaction record characteristic value, interaction strength and distribution thereof, and can be specifically expressed as follows:

CI＝{R_in/out，C_out，C_in，C_in/out，F_in/out}

wherein R is_in/outAs a proportion of subscribers as calling and called subscribers, C_outThe number of outgoing calls a subscriber has in all calls with different subscribers only as a calling subscriber, C_inThe number of the user who has the answering action only as the called user in all the calls with different users, C_in/outIs the number of subscribers simultaneously acting as calling and called subscribers in all calls with different subscribers, F_in/outIs the interaction intensity frequency distribution with other users;

(2) establishing call frequency and distribution thereof F_CDThe feature for describing the calling time in the call record includes two parts, namely a calling frequency value in a specified statistical time period and a calling frequency value time distribution, which can be specifically expressed as:

<math> <mrow> <msub> <mi>F</mi> <mi>CD</mi> </msub> <mo>=</mo> <mo>{</mo> <msubsup> <mi>F</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> <mi>T</mi> </msubsup> <mo>,</mo> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mi>T</mi> </msubsup> <mo>}</mo> </mrow> </math>

wherein,

to count the absolute frequency value of calls made by users within a time period,

is the distribution of the frequency of calls of the users within a statistical time period over 12 time slices of 2 hours in length a day.

(3) Set-up call duration and distribution thereof D_CTFeatures for describing duration in call records, including call rejection ratio, call average in statistical time periodsThe histogram distribution of the mean duration and the call duration may be specifically expressed as:

$D_{\mathrm{CT}}=\{f_E^T,{\mathrm{CT}}_{\mathrm{avg}}^T,{\mathrm{CTD}}^T\}$

wherein,

a probability value for refusal of answering in the call initiated by the user,

average call duration, CTD, for a subscriber^TIs the distribution of call duration.

2. The generation method according to claim 1, characterized in that step (1) comprises in particular the sub-steps of:

(1-1) statistical user's historical call interaction behavior characteristic parameter C_out，C_inAnd C_in/outIn order to facilitate the lateral comparison among users, the method further performs normalization processing on the user data:

<math> <mrow> <msub> <mover> <mi>C</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> </msub> <mo>=</mo> <mfrac> <msub> <mi>C</mi> <mi>out</mi> </msub> <mrow> <msub> <mi>C</mi> <mi>out</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mi>in</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> </mrow> </mfrac> </mrow> </math>

<math> <mrow> <msub> <mover> <mi>C</mi> <mo>&OverBar;</mo> </mover> <mi>in</mi> </msub> <mo>=</mo> <mfrac> <msub> <mi>C</mi> <mi>in</mi> </msub> <mrow> <msub> <mi>C</mi> <mi>out</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mi>in</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> </mrow> </mfrac> </mrow> </math>

<math> <mrow> <msub> <mover> <mi>C</mi> <mo>&OverBar;</mo> </mover> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> <mo>=</mo> <mfrac> <msub> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> <mrow> <msub> <mi>C</mi> <mi>out</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mi>in</mi> </msub> <mo>+</mo> <msub> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> </mrow> </mfrac> </mrow> </math>

wherein,and

the sum of the three is 1;

(1-2) statistics of incoming and outgoing call proportion R_in/outThe calculation formula is as follows:

$R_{\mathrm{in}/\mathrm{out}}=\frac{C_{\mathrm{out}}}{C_{\mathrm{in}}}$

(1-3) counting the call interaction intensity CD of the specified user and other users_in/out. The strength of the call interaction between the user and user j is expressed as

The calculation formula is as follows:

${\mathrm{CD}}_{\mathrm{in}/\mathrm{out}}^j=\mathrm{INT}\left[\log (C_{\mathrm{out}}^j+C_{\mathrm{in}}^j)\right]$

wherein INT [. cndot]The function of rounding is represented by a number of,

representing the number of times the user actively calls user j,

indicating the number of times the user answered the call from user j;

(1-4) statistical Call interaction Strength distribution CDD_in/out

CDD_in/out＝{CDN₀，CDN₁，CDN₂，CDN₃}

Among them, CDN_iThe calculation formula of (i ═ 0,1, 2, 3) is as follows:

CDN (content delivery network)_iIs DC of value equal to i or greater than or equal to i_in/outOf wherein COUNT [ ·]In order to be a function of the count,

representing the call interaction strength between the user and the user j, and n representing the number of all contacts of the user;

(1-5) for CDD_in/_outNormalization processing is carried out, and the call interaction intensity distribution after normalization is recorded asThe concrete expression is as follows:

<math> <mrow> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> </msub> <mo>=</mo> <mo>{</mo> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mn>0</mn> </msub> <mo>,</mo> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mn>1</mn> </msub> <mo>,</mo> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mn>2</mn> </msub> <mo>,</mo> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mn>3</mn> </msub> <mo>}</mo> </mrow> </math>

wherein,

for normalized CDN_i(i ═ 0,1, 2, 3), the calculation formula is:

<math> <mrow> <msub> <mover> <mi>CDN</mi> <mo>&OverBar;</mo> </mover> <mi>i</mi> </msub> <mo>=</mo> <mfrac> <msub> <mi>CDN</mi> <mi>i</mi> </msub> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>k</mi> <mo>+</mo> <mn>0</mn> </mrow> <mi>a</mi> </msubsup> <msub> <mi>CDN</mi> <mi>k</mi> </msub> </mrow> </mfrac> <mo>.</mo> </mrow> </math>

3. the generation method according to claim 1, characterized in that step (2) comprises in particular the sub-steps of:

(2-1) counting the absolute frequency value of the call initiated by the user

$F_{\mathrm{in}/\mathrm{out}}^T=\frac{C_{\mathrm{out}}^T}{T}$

Wherein, T is a designated statistical time period length, and the unit is hour, the time period should start from 0 of the first day and end at 24 of the last day, so T should be an integral multiple of 24 hours;

the number of calling times of the user as the calling user in the time period;

(2-2) counting call frequency distribution parameters of a specified subscriber within a specified time period T

The concrete expression is as follows:

$D_{\mathrm{out}}^T=\{D_{\mathrm{out}}^1,D_{\mathrm{out}}^2,...,D_{\mathrm{out}}^{12},\}$

wherein,represents the sum of the number of calls in the [2 x (T-1),2 x T) period of each day of the statistical time period T. Each day time can be divided into 12 time slices, each time slice comprises 2 hours, and the 12 time slices specifically comprise: [0,2),[2,4),[4,6),[6,8),[8,10),[10,12),[12,14),[14,16),[16,18),[18,20),[20,22),[22,24).

(2-3) Call frequency distribution parameter to subscribers

Go on to unityThe chemical treatment process comprises the following steps:

<math> <mrow> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mi>t</mi> </msubsup> <mo>=</mo> <mfrac> <msubsup> <mi>D</mi> <mi>out</mi> <mi>t</mi> </msubsup> <msubsup> <mi>C</mi> <mi>out</mi> <mi>T</mi> </msubsup> </mfrac> <mo>,</mo> <mrow> <mo>(</mo> <mi>t</mi> <mo>=</mo> <mn>1</mn> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>.</mo> <mo>,</mo> <mn>12</mn> <mo>)</mo> </mrow> <mo>;</mo> </mrow> </math>

the normalized call frequency distribution is then:

<math> <mrow> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mi>T</mi> </msubsup> <mo>=</mo> <mo>{</mo> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mn>1</mn> </msubsup> <mo>,</mo> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mn>2</mn> </msubsup> <mo>,</mo> <mo>.</mo> <mo>.</mo> <mo>.</mo> <mo>,</mo> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mn>12</mn> </msubsup> <mo>,</mo> <mo>}</mo> <mo>;</mo> </mrow> </math>

(2-4) obtaining the calling frequency with use efficiency and the distribution parameter F thereof_CD

<math> <mrow> <msub> <mi>F</mi> <mi>CD</mi> </msub> <mo>=</mo> <mo>{</mo> <msubsup> <mi>F</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> <mi>T</mi> </msubsup> <mo>,</mo> <msubsup> <mover> <mi>D</mi> <mo>&OverBar;</mo> </mover> <mi>out</mi> <mi>T</mi> </msubsup> <mo>}</mo> <mo>.</mo> </mrow> </math>

4. The generation method according to claim 1, characterized in that step (3) comprises in particular the sub-steps of:

(3-1) counting the rate of refused answering in the calling initiated by the user

The calculation formula is as follows:

$f_E^T=\frac{{\mathrm{CR}}_{\mathrm{out}}^T}{C_{\mathrm{out}}^T}$

wherein,representing all calls made by the user within a statistical time period,

representing the rejected active call times in the statistical period of time;

(3-2) counting average call duration of the subscriber

The calculation formula is as follows:

<math> <mrow> <msubsup> <mi>CT</mi> <mi>avg</mi> <mi>T</mi> </msubsup> <mo>=</mo> <mfrac> <mrow> <msubsup> <mi>&Sigma;</mi> <mrow> <mi>r</mi> <mo>=</mo> <mn>1</mn> </mrow> <msubsup> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> <mi>T</mi> </msubsup> </msubsup> <msub> <mi>t</mi> <mi>r</mi> </msub> </mrow> <msubsup> <mi>C</mi> <mrow> <mi>in</mi> <mo>/</mo> <mi>out</mi> </mrow> <mi>T</mi> </msubsup> </mfrac> </mrow> </math>

wherein,

representing the number of calls in a statistical period of time, t_rRecording the corresponding call duration for the r-th call;

(3-3) histogram distribution of statistical Call duration CTD^TSpecifically, the following are shown:

${\mathrm{CTD}}^T=\{{\mathrm{CT}}_D^0,{\mathrm{CT}}_D^1,...,{\mathrm{CT}}_D^6,{\mathrm{CT}}_D^7\}$

wherein,

the call record proportion of the call duration in a certain time period range is represented, and the specific calculation formula is as follows:

Images