CN107563402A - A kind of social networks estimating method and system - Google Patents

Abstract

The present invention, which provides a kind of social networks estimating method and system, methods described, to be included：Based on tensor resolution method, obtained from space-time data in the same period, multiple mobile intentions of the multiple mobile intentions and second user of the first user；Based on default multi classifier, from multiple mobile intentions of the multiple mobile intentions and second user of first user, the mobile intention of first user and the second user co-occurrence is obtained；Mobile intention based on the disaggregated models of SVM bis- and the co-occurrence, infer the social networks of first user and the second user.Social networks estimating method provided by the invention and system, being decomposited by tensor resolution method has the mobile intention of same space-time characterisation in user's space-time data, so as to establish the mapping relations between space-time data and the mobile intention, and then infer the social relationships between user, improve the accuracy rate of deduction.

Description

A social relationship inference method and system

technical field

The present invention relates to the field of information technology, and more specifically, to a social relationship inference method and system.

Background technique

At present, with the widespread popularization of smart mobile devices with various sensors such as mobile phones and tablet computers, and the rapid development of social networks, more and more spatio-temporal data can be collected, such as: personal communication data, GPS trajectory, different location services The sign-in data provided by the provider, etc. And with the development of the Internet of Things, systems originally used for other purposes have further increased the ability to collect spatio-temporal data, such as smart bus card systems, video surveillance systems, ATM systems, etc. Mining the social relationship between people from these spatio-temporal data is crucial for some popular applications, such as friend recommendation, advertisement placement, spread of epidemics, identification of members of criminal or terrorist organizations, etc.

In order to infer the social relationship between two people from the spatio-temporal data, the prior art uses the co-occurrence characteristics of the two people to infer the closeness of the relationship between the two people. Co-occurrence of two people refers to the phenomenon that two people appear in the same place at the same time. Generally speaking, the closer the relationship between two people, the more times they co-occur, so the closeness of the relationship between two people can be judged directly by the number of co-occurrences. The co-occurrence location can be determined as a public place or a private place according to the entropy of the location. The calculation formula of the location entropy is as follows:

Among them, P _u,l represents the ratio of the number of times a certain user appears in the lth location to the total number of times that all users appear in the lth location. If the entropy of the location is greater, it means that many people have been to the location, and the location is a public place. Then the co-occurrence of locations with higher entropy contributes less to the relationship between the two, while the co-occurrence of locations with lower entropy contributes more to the relationship between the two, then the amount of co-occurrence can be inferred from the entropy value, so that Infer the relationship between two people.

However, two closely related people will also co-occur in places with high entropy, such as large shopping malls, supermarkets, etc., and most of the existing spatio-temporal data come from public places, so the spatio-temporal data of places with low entropy It is not easy to obtain, so in practical applications, the existing methods of inferring social relationships based on entropy values are not accurate.

Contents of the invention

The present invention provides a social relationship inference method and system to overcome the above problems or at least partially solve the above problems.

According to the first aspect provided by the present invention, the present invention provides a social relationship inference method, the method comprising:

S1. Based on the tensor decomposition method, multiple mobile intentions of the first user and multiple mobile intentions of the second user are obtained from the spatio-temporal data within the same time period;

S2. Based on a preset multi-class classifier, from the multiple mobile intentions of the first user and the multiple mobile intentions of the second user, obtain the co-occurring mobile intentions of the first user and the second user ;

S3. Based on the SVM binary classification model and the co-occurring mobile intention, infer the social relationship between the first user and the second user.

Wherein, step S1 includes:

dividing the spatio-temporal data into multiple spatio-temporal sub-data according to a preset time period;

Calculating the tensor element values of the spatiotemporal sub-data in each time period;

Based on the eigenvectors of the tensor element values, a plurality of mobile intentions of the first user and a plurality of mobile intentions of the second user are extracted.

Wherein, the method described before step S1 also includes:

The spatio-temporal data is obtained, and part of the spatio-temporal data is used as a training data set to train the multi-class classifier.

Wherein, the acquisition of the spatio-temporal data and using part of the spatio-temporal data as a training data set to train the multi-class classifier includes:

determining classification features for the multi-class classifier based on the tensor element values;

Based on the classification features, the multi-class classifier is trained using the training data set.

Wherein, step S2 includes:

calculating a classification feature value of the first user based on a plurality of movement intentions of the first user;

calculating a classification feature value of the second user based on a plurality of movement intentions of the second user;

Based on the multi-class classifier, comparing the classification feature value of the first user with the classification feature value of the second user, extracting the co-occurring movement intention of the first user and the second user.

Wherein, the classification features include:

One or more of spatial features, time features, and date features.

Wherein, step S3 includes:

converting the co-occurring mobile intentions of the first user and the second user into a multidimensional feature vector;

Based on the SVM binary classification model and the multi-dimensional feature vector, infer the social relationship between the first user and the second user.

According to a second aspect of the present invention, a social relationship inference system is provided, including:

The tensor decomposition module is used to obtain multiple mobile intentions of the first user and multiple mobile intentions of the second user within the same time period from the spatio-temporal data based on the tensor decomposition method;

A multi-class classifier module, configured to obtain the first user and the second user from the multiple mobile intentions of the first user and the multiple mobile intentions of the second user based on a preset multi-class classifier. co-occurring mobile intent of users;

An inference module, configured to infer the social relationship between the first user and the second user based on the SVM binary classification model and the co-occurring mobile intention.

According to a third aspect of the present invention, a computer program product is provided, including program code, the program code is used to execute the above-mentioned social relationship inference method.

According to a fourth aspect of the present invention, there is provided a non-transitory computer-readable storage medium for storing the aforementioned computer program.

The social relationship inference method and system provided by the present invention decompose the mobile intentions with the same spatiotemporal characteristics in the user's spatiotemporal data through the tensor decomposition method, so as to establish the mapping relationship between the spatiotemporal data and the mobile intentions, and then infer the relationship between users. social relations, which improves the accuracy of inference.

Description of drawings

Fig. 1 is a flow chart of a social relationship inference method provided by an embodiment of the present invention;

Fig. 2 is a comparison chart of the experimental results of the social relationship inference method provided by the embodiment of the present invention;

Fig. 3 is a structural diagram of a social relationship inference system provided by an embodiment of the present invention.

detailed description

The specific implementation manners of the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

Fig. 1 is a flow chart of a social relationship inference method provided by an embodiment of the present invention, as shown in Fig. 1 , including:

S1. Based on the tensor decomposition method, multiple mobile intentions of the first user and multiple mobile intentions of the second user are obtained from the spatio-temporal data within the same time period;

S2. Based on a preset multi-class classifier, from the multiple mobile intentions of the first user and the multiple mobile intentions of the second user, obtain the co-occurring mobile intentions of the first user and the second user ;

S3. Based on the SVM binary classification model and the co-occurring mobile intention, infer the social relationship between the first user and the second user.

In the S1, the tensor decomposition method is a CP decomposition method, that is, a tensor is expressed as a sum of finite rank tensors, and the CP decomposition method can effectively decompose the spatiotemporal characteristics in the spatiotemporal data, based on The tensor decomposition method can extract the user's movement pattern from spatio-temporal data, that is, the user's movement intention in different time periods, and the movement intention is the movement purpose.

In the above S2, it can be understood that the multi-class classifier is used to establish a mapping function between user records and mobile purposes, and when there are multiple mobile purposes, the classification of mobile purposes can be regarded as A multi-class classification, therefore, a multi-class classifier is set up to determine the desired characteristics of the mobile purpose.

In the S3, the SVM binary classification model is a traditional classifier, which is a very effective linear classifier. According to the co-occurring destination information, the linear classifier can be used to infer whether the first user and the second user belong to strangers or acquaintance category.

The social relationship inference method and system provided by the present invention decompose the mobile destinations with the same spatio-temporal characteristics in the user's spatio-temporal data through the tensor decomposition method, so as to establish the mapping relationship between the spatio-temporal data and the mobile destinations, and then infer the user The social relationship between them improves the accuracy of the calculation.

On the basis of the foregoing embodiments, step S1 includes:

dividing the spatio-temporal data into multiple spatio-temporal sub-data according to a preset time period;

Calculating the tensor element values of the spatiotemporal sub-data in each time period;

Based on the eigenvectors of the tensor element values, a plurality of mobile intentions of the first user and a plurality of mobile intentions of the second user are extracted.

Specifically, for example: the data obtained is the data of the Beijing Public Transport Card, which collects the card swiping records of major buses in the urban area of Beijing from October 1, 2014 to October 31, 2014.

Using the tensor decomposition algorithm, the moving purpose in the data set is extracted. Calculate each tensor element The value of is calculated as follows:

Take all the unique bus stops in the data as locations, corresponding to a r _i . Two bus stops are considered one bus stop if they are no more than 300 meters apart. Divide a day into the following eight time periods, and the time of each time period is as follows:

00:00:00-4:59:59, 05:00:00-6:59:59, 07:00:00-8:59:59, 09:00:00-10:59:59, 11: 00:00-13:59:59, 14:00:00-16:59:59, 17:00:00-19:59:59, 20:00:00-23:59:59.

Count the number of all passengers in each time period above on October 1st and October 31st And the number of passengers Count(ri , _{t j ,} d _k ) of each bus station r _i , then each tensor element The value is:

After obtaining the values of all tensor elements, use the CP decomposition procedure to perform CP decomposition on the tensor Y:

Among them, λ _r is the coefficient, and Y _r is the obtained first-order tensor, which can be written as the outer product of three vectors a _r , b _r , and c _r .

Among them, a _r , b _r , and c _r are the feature vectors of location, time and date respectively; and Y _r is a mobile destination we extracted from the data set. On the data set of the Beijing Public Transport Card we used, we extracted 7 kinds of mobile purposes.

It should be noted that the seven moving purposes extracted from the card swiping record data of major buses in Beijing from October 1, 2014 to October 31, 2014 provided by the embodiment of the present invention are only the inherent moving purposes in the scheme , the embodiment of the present invention does not specifically limit the number of mobile purposes in different data sets.

The embodiment of the present invention uses tensor decomposition method to extract the user's movement purpose in the spatio-temporal data set as a discriminant feature, which is easy to obtain and has obvious features, and provides a discriminative basis for subsequent social relationship inference between users.

On the basis of the foregoing embodiments, the method before step S1 further includes:

The spatio-temporal data is obtained, and part of the spatio-temporal data is used as a training data set to train the multi-class classifier.

Specifically, we collected the spatio-temporal data we obtained, and identified 412 cardholders according to the recorded time and place, and determined 2796 pairs of known users through investigation, and used the existing algorithm to extract the 412 cardholders. All co-occurrences among users, the co-occurrences include all co-occurrences between known and unknown users. We randomly selected 80% of the 2,796 pairs of users we knew as training data, and trained a multi-class classifier to establish a mapping function between user records and mobile purposes.

On the basis of the foregoing embodiments, the acquisition of the spatio-temporal data, and using a partial data set of the spatio-temporal data as a training data set to train the multi-class classifier includes:

determining classification features for the multi-class classifier based on the tensor element values;

Based on the classification features, the multi-class classifier is trained using the training data set.

Based on the tensor element values of the tensor decomposition in the S1 process, the tensor information can be divided into multiple types of features, so which type of movement purpose each record corresponds to is equivalent to which type of movement purpose the corresponding mapping is in, Classify them according to their characteristics.

On the basis of the foregoing embodiments, step S2 includes:

calculating a classification feature value of the first user based on a plurality of movement intentions of the first user;

calculating a classification feature value of the second user based on a plurality of movement intentions of the second user;

Based on the multi-class classifier, comparing the classification feature value of the first user with the classification feature value of the second user, extracting the co-occurring movement intention of the first user and the second user.

Specifically, the embodiment of the present invention uses the existing Adaboost algorithm and training data set to train the Adaboost model. According to the trained Adaboost classification model, these co-occurring destination information are mapped into mobile destination pairs. In the embodiment of the present invention, seven kinds of moving purposes are extracted, and seven different moving purpose pairs are generated, and the moving purpose pairs are the extracted co-occurrence features.

On the basis of the foregoing embodiments, the classification features include:

One or more of spatial features, time features, and date features.

According to the information given by the first-order tensor decomposed before, and some feature engineering experiments we have done, we have determined three major types of features, namely spatial features, time features, and date features. The classification features provided by the embodiments of the present invention One or more of these three types of features may be included, and these three types of features specifically include:

Spatial features: wherein, the spatial features include location entropy, which is used to measure the heat of a location, and its calculation formula is as follows:

where p _i is the ratio of the number of times that the i-th user appears in lock _k to the number of times that all users appear in lock _k .

The spatial feature also includes a location type, which indicates the category to which the location belongs, such as a bar, a supermarket, and the like. The location type can be obtained through some application programming interfaces based on location services.

The spatial feature also includes a distance, which indicates the distance between the location in the current record and the location in the previous record.

Time characteristic: Wherein, the time characteristic includes hour, and the hour represents the recorded time, and the value ranges from 0 to 23.

The time feature also includes a pseudo-stay time, which represents the time difference between the time in the current record and the time in the previous record.

The time feature also includes a time interval, and the time interval is used to measure a certain location and a certain user's access habit. Equal to the difference between the time in the current record and the time in the previous record for the same location.

The time feature also includes a secondary time interval, and the secondary time interval is also used to measure a certain location and a certain user's access habit. Equal to the difference between the time in the current record and the time in the second previous record for the same location.

Date feature: Wherein, the date feature includes a week, and the week uses 0 to 6 to represent Sunday, Monday to Saturday.

The date feature also includes a date, and the date uses 0 to 31 to represent the number of days in the month.

The date feature also includes a date type, and the date type represents a date category. The embodiment of the present invention provides three date categories: weekends, small holidays and long holidays, corresponding to two-day holidays on weekends and three-day holidays such as the Mid-Autumn Festival. holidays up to five days, and holidays exceeding five days or more.

The embodiment of the present invention refines the classification by specifically classifying the classification features into spatial features, time features, and date features, so that the classification is more accurate.

On the basis of the foregoing embodiments, step S3 includes:

converting the co-occurring mobile intentions of the first user and the second user into a multidimensional feature vector;

Based on the SVM binary classification model and the multi-dimensional feature vector, infer the social relationship between the first user and the second user.

Specifically, for example, the obtained co-occurring movement intentions of the first user and the second user are the data provided in the above embodiment, and the number of information to be calculated is 7, and a different mobile destination pairs. Counting all mobile destination pairs of two users to form a 21-dimensional feature vector, using the 21-dimensional feature vector to train the SVM binary classification model, and then judging whether the first user and the second user are based on the 21-dimensional feature vector Strangers or acquaintances.

Fig. 2 is a comparison chart of the inference results of the social relationship inference method provided by the embodiment of the present invention. As shown in Fig. 2, the black solid line is the inference result of the embodiment of the present invention, and the black dotted line is the inference result of the prior art. From Fig. 2 It can be seen that in terms of inference accuracy, the embodiment of the present invention is significantly better than the method inference based on the co-occurrence of location entropy in the prior art.

The social relationship inference method and system provided by the present invention decompose the mobile destinations with the same spatio-temporal characteristics in the user's spatio-temporal data through the tensor decomposition method, so as to establish the mapping relationship between the spatio-temporal data and the mobile destinations, and then infer the user The social relationship between them improves the accuracy of the calculation.

FIG. 3 is a structural diagram of a social relationship inference system provided by an embodiment of the present invention, including: a tensor decomposition module 1, a multi-class classifier module 2, and an inference module 3, wherein,

The tensor decomposition module 1 is used to obtain multiple mobile intentions of the first user and multiple mobile intentions of the second user within the same time period from the spatio-temporal data based on the tensor decomposition method;

The multi-class classifier module 2 is used to obtain the first user and the second user from the multiple mobile intentions of the first user and the multiple mobile intentions of the second user based on a preset multi-class classifier. co-occurring mobile intent of users;

The inference module 3 is used to infer the social relationship between the first user and the second user based on the SVM binary classification model and the co-occurring mobile intention.

For a specific social relationship inference method, reference may be made to the foregoing embodiments, and details are not repeated in this embodiment of the present invention.

This embodiment provides a social relationship inference system, including: at least one processor; and at least one memory communicated with the processor, wherein:

The memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to execute the methods provided by the above method embodiments, for example, including: S1, based on the tensor decomposition method, from the space-time Obtain multiple mobile intentions of the first user and multiple mobile intentions of the second user within the same time period from the data; S2, based on a preset multi-class classifier, from the multiple mobile intentions of the first user and the second Among the multiple mobile intentions of two users, obtain the co-occurring mobile intentions of the first user and the second user; S3, based on the SVM binary classification model and the co-occurring mobile intentions, infer the first user and the co-occurring mobile intentions The social relationship of the second user.

This embodiment discloses a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, and when the program instructions are executed by the computer, the computer The methods provided by the above-mentioned method embodiments can be executed, for example, including: S1. Based on the tensor decomposition method, obtain multiple mobile intentions of the first user and multiple mobile intentions of the second user within the same time period from the spatio-temporal data ; S2, based on a preset multi-class classifier, from the multiple mobile intentions of the first user and the multiple mobile intentions of the second user, obtain the co-occurring movement of the first user and the second user Intent; S3. Based on the SVM binary classification model and the co-occurring mobile intention, infer the social relationship between the first user and the second user.

This embodiment provides a non-transitory computer-readable storage medium, the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the methods provided in the above method embodiments, for example, including : S1. Based on the tensor decomposition method, multiple mobile intentions of the first user and multiple mobile intentions of the second user are obtained from the spatiotemporal data in the same time period; S2. Based on the preset multi-class classifier, from the Among the multiple mobile intentions of the first user and the multiple mobile intentions of the second user, obtain the co-occurring mobile intentions of the first user and the second user; S3, based on the SVM binary classification model and the co-occurrence mobile intention, inferring the social relationship between the first user and the second user.

Those of ordinary skill in the art can understand that all or part of the steps for realizing the above-mentioned method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the It includes the steps of the above method embodiments; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk and other various media that can store program codes.

Through the above description of the implementations, those skilled in the art can clearly understand that each implementation can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the prior art can be embodied in the form of software products, and the computer software products can be stored in computer-readable storage media, such as ROM/RAM, magnetic discs, optical discs, etc., including several instructions to make a computer device (which may be a personal computer, server, or network device, etc.) execute the methods described in various embodiments or some parts of the embodiments.

Finally, the method of the present application is only a preferred embodiment, and is not intended to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Claims (10)

1. A social relationship inference method, characterized in that, comprising: S1. Based on the tensor decomposition method, multiple mobile intentions of the first user and multiple mobile intentions of the second user are obtained from the spatio-temporal data within the same time period; S2. Based on a preset multi-class classifier, from the multiple mobile intentions of the first user and the multiple mobile intentions of the second user, obtain the co-occurring mobile intentions of the first user and the second user ; S3. Based on the SVM binary classification model and the co-occurring mobile intention, infer the social relationship between the first user and the second user. 2. The method according to claim 1, wherein step S1 comprises: dividing the spatio-temporal data into multiple spatio-temporal sub-data according to a preset time period; Calculating the tensor element values of the spatiotemporal sub-data in each time period; Based on the eigenvectors of the tensor element values, a plurality of mobile intentions of the first user and a plurality of mobile intentions of the second user are extracted. 3. The method according to claim 2, characterized in that, the method before step S1 also includes: The spatio-temporal data is obtained, and part of the spatio-temporal data is used as a training data set to train the multi-class classifier. 4. method according to claim 3, is characterized in that, described acquisition described spatio-temporal data, and with the part data of described spatio-temporal data as training data set, described multiclass classifier is trained, comprising: determining classification features for the multi-class classifier based on the tensor element values; Based on the classification features, the multi-class classifier is trained using the training data set. 5. The method according to claim 4, wherein step S2 comprises: calculating a classification feature value of the first user based on a plurality of movement intentions of the first user; calculating a classification feature value of the second user based on a plurality of movement intentions of the second user; Based on the multi-class classifier, comparing the classification feature value of the first user with the classification feature value of the second user, extracting the co-occurring movement intention of the first user and the second user. 6. The method according to claim 5, wherein the classification features include: One or more of spatial features, time features, and date features. 7. The method according to any one of claims 1-6, wherein step S3 comprises: converting the co-occurring mobile intentions of the first user and the second user into a multidimensional feature vector; Based on the SVM binary classification model and the multi-dimensional feature vector, infer the social relationship between the first user and the second user. 8. A social relationship inference system, characterized in that it comprises: The tensor decomposition module is used to obtain multiple mobile intentions of the first user and multiple mobile intentions of the second user within the same time period from the spatio-temporal data based on the tensor decomposition method; A multi-class classifier module, configured to obtain the first user and the second user from the multiple mobile intentions of the first user and the multiple mobile intentions of the second user based on a preset multi-class classifier. co-occurring mobile intent of users; An inference module, configured to infer the social relationship between the first user and the second user based on the SVM binary classification model and the co-occurring mobile intention. 9. A computer program product, characterized in that the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, and when the program instructions are executed by a computer , causing the computer to execute the method according to any one of claims 1-7. 10. A non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the computer according to any one of claims 1 to 7. Methods.