A method for identifying family relationships using mobile phone signaling data
Through the preprocessing and feature extraction of mobile phone signaling data, combined with decision tree and KNN algorithm, family relationships are identified, and the problems of long investigation cycles and small ranges in the existing technology are solved, and the rapid and comprehensive family relationship recognition and commuting space are achieved. Analysis provides an effective basis for urban planning.
Patent Information
- Application Number
- CN202210217274.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-06
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-03-06
AI Technical Summary
When obtaining and analyzing traffic patterns and commuting space characteristics, the prior art has problems such as long investigation periods, small scopes and huge resource consumption, making it difficult to effectively use mobile phone signaling data for family relationship identification.
By defining mobile phone signaling data sets, data preprocessing, residence location identification, residence location distance comparison, call feature indicator extraction and decision tree classification are carried out, combined with KNN algorithm, family relationships are accurately identified, and the defects of manual investigation are compensated for, and urban planning is provided. in accordance with.
It realizes rapid and comprehensive family relationship identification, with a large coverage, continuous time and complete space, supports urban commuting space characteristics analysis and planning, and reduces resource consumption.
Smart Images

Figure CN114501420B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of transportation technology. The mobile phone signaling data is used to accurately infer the family relationship between mobile phone users, which can be further used for commuting space characteristics, travel pattern analysis, etc. Background Art
[0002] Households are an important unit structure for analyzing traffic patterns or commuting space characteristics. This data is often obtained through manual means such as population censuses or traffic surveys, such as resident travel surveys. Although manual surveys are relatively comprehensive, they have disadvantages such as long survey cycles, small survey scopes, and huge resource consumption.
[0003] Mobile phone signaling data is the data generated when the user's mobile phone connects to the nearby base station when communicating with the outside world. It has the advantages of temporal continuity, spatial integrity, wide coverage, and high granularity, which makes up for the defects of long period and small survey scope in manual survey. Mobile phone signaling data is used for analysis to obtain the family relationship between users in the signaling data, providing a basis for the analysis of work-residence space characteristics and the formulation of urban planning schemes. Mobile phone signaling data contains information such as user unique identification code, timestamp, base station number, longitude, latitude, and call object, as shown in Table 1.
[0004] Table 1: Mobile phone signaling data information
[0005] Summary of the invention
[0006] The present invention proposes a method for accurately identifying family relationships using mobile phone signaling data. The identification of family relationships can make up for the long cycle defects in population census and traffic survey methods. At the same time, the family relationships can support urban commuting space feature analysis, etc., and provide a basis for urban development planning.
[0007] The specific implementation method is as follows:
[0008] Define the total mobile phone signaling data set to be calculated as U = {U1, U2…U i},U i represents the mobile phone signaling dataset of user i, U i = {u i,1 ,u i,2 …u i,j},u i,j represents the jth mobile phone signaling data of user i, u i,j ={U_ID i ,T i,j ,LON i,j ,LAT i,j ,AREA i,j ,CU i,j}, where U_IDi , T i,j ,LON i,j , LAT i,j 、AREA i,j , CU i,j They respectively represent the user unique identification code of the jth mobile phone signaling data of user i, the timestamp of the data generation, the longitude of the communication base station, the latitude of the communication base station, the traffic community to which the base station belongs, and the communication object code.
[0009] The specific steps are as follows:
[0010] Step 1: Data preprocessing: Define the night time period as 20:00:00-23:59:59 of the current day and 00:00:00-7:00:00 of the next day, and i The signaling data is sorted according to the time of event occurrence, and T i,j The data recorded in the night range is matched with the corresponding night time period according to the defined night sequence and the time of the user signaling data to construct the night data set B = {B i,1 ,B i,2 …B i,m}, B i,m Indicates that user U in the mth night time period i A collection of mobile phone signaling data;
[0011] Step 2: Residential location identification: Calculate user U i The night residence of B, in the mth night, the user connects to the traffic cell AREA to which the base station belongs i,j The length of stay of user U is calculated by accumulating the night stay time of the traffic cells belonging to the same base station within a certain period, and selecting the traffic cell with the largest stay time as the user U i The place of residence is and They represent the longitude and latitude of the place of residence, i.e., the centroid coordinates of the traffic cell to which the base station with the longest nighttime stay belongs;
[0012] Step 3: Comparison of residential distance: For all other users Q = {Q1, Q2, ..., Q n}, i≠n, n is a natural number, and n>0, the place of residence is determined, that is, the operations in steps 1 and 2 are repeated, and the set Q and the user U are calculated respectively. i Straight-line distance between residences in For user Q n The residential location information of user Q nThe centroid coordinates of the traffic cell to which the base station with the longest nighttime stay belongs are taken out, and the users whose elements in D are less than 1 km away are formed to form the data set Z of the user, and at the same time, the user pair set is constructed. Where v is a natural number and v>0, n is a natural number and n>0, Refers to user Q v With user U i The user pairs connected by the unique identification code; users whose elements in D are greater than or equal to 1KM are considered to have no family relationship;
[0013] Step 4: Extract call feature indicators and calculate the set QU i The characteristic index of each element in The characteristic indicators include the following four indicators:
[0014] 1) Number of days with the same residence at night: During the night, users with family relationships tend to have the same residence compared to non-family members. Therefore, we propose the number of days with the same residence at night, denoted as That is, within the T period, the user The number of days that the nighttime residence is the same;
[0015] 2) Number of days of calls: Compared with users who are not family members, the calling behavior between family members is more stable. We propose the indicator “number of days of calls”, recorded as That is, users The number of days during which calling behavior occurs within the T period;
[0016] 3) Night call frequency: Due to the co-living behavior between family members, the frequency of night calls is less, which is significantly different from non-family users. The "night call frequency" indicator is proposed, that is, The calculation formula is as follows:
[0017]
[0018] in, Refers to user Q v With user U i The total number of calls at night during the T period, Refers to QU i Each user The sum of the number of calls at night during period T;
[0019] 4) Evening peak call frequency: The evening peak period is defined as 17:00-19:00, and the "evening peak call frequency" indicator is proposed, that is, Refers to the user The call frequency during the evening peak period in period T:
[0020]
[0021] in, It means that within the T period, user U i Q with user v Total number of calls during the evening peak period, Refers to the user The total number of calls in period T;
[0022] For QU i Each user Constructing feature sets
[0023] Step 5: Use the classification rules of the decision tree to remove users who are unlikely to have a family relationship, based on the number of days in which the nighttime residence is the same in the feature set P. and call days Two indicators are used to judge. When the number of days that two users have the same nighttime residence is 0, it is impossible for the two users to have a family relationship in terms of residence. When the number of days that the two users have the same nighttime residence is greater than 0, if the number of days that the two users make calls within a certain period is greater than 0, it is considered that the two users "may have a family relationship", otherwise it is not a family relationship. Based on this, the feature set of user pairs that "may have a family relationship" is extracted. f≤v, f is a natural number, and f>0, v is a natural number, and v>0, G f U i Refers to the user
[0024] Step 6: Family relationship classification based on KNN algorithm: Based on the known family user pairs and non-family user pairs in the artificial samples, the user pair feature set W is constructed using the same method, and the relationship between the user pairs is calibrated. 1≤a≤n, n is a natural number, and n>0, S a H a Refers to the user L aa Refers to the user The family relationship label between them, the feature set for users who “may have family relationships” f≤v, f is a natural number, and f>0, v is a natural number, and v>0, and the KNN algorithm is further used for classification. The process is as follows:
[0025] (1) First, calculate the feature set of users who may have family relationships f is a natural number, and f>0, v is a natural number, and v>0, and the set W of known labels The Euclidean distance between them is calculated as Construct the Euclidean distance set E = {E1, E2, ..., E a}, E a Indicates the user With users The Euclidean distance between
[0026] (2) Sort the Euclidean distance set E in ascending order according to the distance value;
[0027] (3) Select the first three samples, determine the frequency of occurrence of the categories to which the first three samples belong, and output the category with the highest frequency;
[0028] Step 7: Family relationship identification: When the highest frequency category among the three samples is family relationship, the user If the highest frequency category in the current three samples is non-family relationship, the user Non-family relationship.
[0029] The present invention has the following beneficial effects:
[0030] The basic data of this invention comes from mobile phone signaling data. Compared with traditional manual investigation methods, this method does not require additional collection equipment. It has the characteristics of covering a large population range, being continuous in time and complete in space, and provides all-round support for exploring family relationships. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 .Night range diagram
[0032] Figure 2 . Decision tree process diagram
[0033] Figure 3 . Schematic diagram of the process DETAILED DESCRIPTION
[0034] The specific implementation process of the present invention is described in detail below in conjunction with the accompanying drawings. Take the mobile phone signaling data of a certain region as an example, as shown in Table 2.
[0035] Table 2: Mobile phone signaling data
[0036]
[0037]
[0038] The specific steps are as follows:
[0039] Step 1: Data preprocessing: Define the night time period as 20:00:00-23:59:59 of the current day and 00:00:00-7:00:00 of the next day, and iThe signaling data is sorted according to the time of event occurrence, and T i,j Data recording during the night time period is based on the defined night sequence (see Appendix Figure 1 ), according to the time of user signaling data, the data is matched with the corresponding night time period, and a night data set B = {B i,1 ,B i,2 …B i,m}, B i,m Indicates that user U in the mth night time period i The set of mobile phone signaling data is composed of, taking user U1 = "3916" as an example, Table 3 shows the first night time period (B 1,1 )’s mobile phone signaling dataset;
[0040] Table 3: Nighttime mobile phone signaling data
[0041]
[0042] Step 2: Residential location identification: Calculate user U i The night residence, with B as the opposite
[0043] For example, during the mth night, the user connects to the traffic cell AREA to which the base station belongs. i,j The length of stay of user U is calculated by accumulating the night stay time of the traffic cells belonging to the same base station within a certain period, and selecting the traffic cell with the largest stay time as the user U i The place of residence is and Respectively represent the longitude and latitude of the place of residence, that is, the centroid coordinates of the traffic cell to which the base station with the longest nighttime stay belongs. Taking user U1 = "3916" as an example, after calculation, the centroid coordinates of the traffic cell to which the base station with the longest stay time belongs (AREA = "4177") are (that is, the place of residence):
[0044] Step 3: Comparison of residential distance: For all other users Q = {Q1, Q2, ..., Q n}i≠n, n is a natural number, and n>0, the place of residence is determined, that is, the operations in steps 1 and 2 are repeated, and the set Q and the user U are calculated respectively. i Straight-line distance between residences in For user Q n The residential location information of user Q n The centroid coordinates of the traffic cell to which the base station with the longest nighttime stay belongs are taken out, and the users whose elements in D are less than 1 km away are formed to form the data set Z of the user, and at the same time, the user pair set is constructed. v≤n, where v is a natural number and v>0, n is a natural number and n>0, Refers to user Q v With user U i The residence of users Q2, Q3, and Q4, i.e., U_ID = "1333", "2924", and "1005", is determined. The centroid coordinates (i.e., residence) of the traffic cells to which the base stations with the longest nighttime stay duration of users Q2, Q3, and Q4 belong are calculated as follows: The straight-line distance between Q2 and U1 is 0.8 km, the straight-line distance between Q3 and U1 is 0 km, and the straight-line distance between Q4 and U1 is 1.9 km. D = {0.8, 0, 1.9}. The user The distance between them does not meet the condition, so the user pair
[0045] Step 4: Extract call feature indicators and calculate user The characteristic indicators include the following four indicators. In this example, T = 7, unit: day:
[0046] 1) The number of days when the night residence is the same: After calculation, the user The number of days with the same nightly residence, User The number of days with the same nightly residence,
[0047] 2) Number of days of calls: calculated
[0048] 3) Night call frequency: calculated separately and Night call frequency indicators:
[0049]
[0050] 4) Evening peak call frequency: by calculating, and The call frequency during the evening peak period is:
[0051]
[0052] For each user pair in QU1 and Constructing feature sets
[0053] Step 5: Use the classification rules of the decision tree to remove users who are unlikely to have family relationships. middle According to the decision tree rule, when the number of days with the same nighttime residence is equal to 0, the two users are not in a family relationship. Belongs to the category of "non-family relationship". Similarly, users Belonging to the category of "possible family relationship", extract the user's The feature set of
[0054] Step 6: Family relationship classification based on KNN algorithm: The feature set W of the known user pair relationship is shown in Table 4 (only part is listed). The KNN algorithm is further used for classification, and the process is as follows:
[0055] (1) Calculation The Euclidean distance between each user pair feature set in the set W of known labels, E = {3.6, 1.4, 3.2, 3.6};
[0056] (2) Sort the Euclidean distance set E in ascending order according to the distance value to E = {1.4, 3.2, 3.6, 3.6};
[0057] (3) Among the categories of the first three samples, the frequency of the “family relationship” category is The frequency of the category "non-family relationship" is
[0058] Step 7: Family relationship determination, due to Therefore, users The relationship is a family relationship.
[0059] Table 4 Feature set W
[0060]
[0061] Finally, it should be noted that the above examples are only used to illustrate the present invention and are not intended to limit the technical solutions described in the present invention. Therefore, although this specification has described the present invention in detail with reference to the above examples, a person skilled in the art should understand that the present invention can still be modified or replaced by equivalents. All technical solutions and improvements that do not depart from the spirit and scope of the invention should be included in the scope of the claims of the present invention.
Claims
1. A method for identifying family relationships using mobile phone signaling data, characterized in that The following steps are involved: Step 1: Data preprocessing: Define the night time period as 20:00:00-23:59:59 of the current day and 00:00:00-7:00:00 of the next day, and i The signaling data is sorted according to the time of event occurrence, and the data records within the night range are extracted. According to the defined night sequence, the data is matched with the corresponding night time period according to the time of user signaling data, and the night data set B = {B i,1 ,B i,2 ...B i,m }, B i,m Indicates that user U in the mth night time period i A collection of mobile phone signaling data; Step 2: Residential location identification: Calculate user U i The night residence of B, in the mth night, the user connects to the traffic cell AREA to which the base station belongs i,j In a cycle, the nighttime stay time of the traffic cells belonging to the same base station is accumulated, and the traffic cell with the largest stay time is selected as the user U i The place of residence is and They represent the longitude and latitude of the place of residence, i.e., the centroid coordinates of the traffic cell to which the base station with the longest nighttime stay belongs; Step 3: Comparison of residential distance: For all other users Q = {Q1, Q2, ..., Q n }, i≠n, n is a natural number, and n>0, the place of residence is determined, that is, the operations in steps 1 and 2 are repeated, and the set Q and the user U are calculated respectively. i Straight-line distance between residences in For user Q n The residential location information of user Q n The centroid coordinates of the traffic cell to which the base station with the longest nighttime stay belongs are taken out, and the users whose elements in D are less than 1 km away are formed to form the data set Z of the user, and at the same time, the user pair set is constructed. Where v is a natural number and v>0, n is a natural number and n>0, Refers to user Q v With user U i The user pairs connected by the unique identification code; users whose elements in D are greater than or equal to 1KM are considered to have no family relationship; Step 4: Extract call feature indicators and calculate the set QU i The characteristic index of each element in The characteristic indicators include the following four indicators: 1) Number of days with the same residence at night: During the night, users with family relationships tend to have the same residence as those with non-family members. Therefore, we propose the number of days with the same residence at night, denoted as That is, within the T period, the user The number of days that the nighttime residence is the same; 2) Number of days of calls: Compared with users who are not family members, the calling behavior between family members is more stable. We propose the indicator "number of days of calls", which is recorded as That is, users The number of days during which calling behavior occurs within the T period; 3) Night call frequency: Due to the co-living behavior between family members, the frequency of night calls is less, which is significantly different from non-family users. Therefore, the "night call frequency" indicator is proposed, that is, The calculation formula is as follows: in, Refers to user Q v With user U i The total number of calls at night during the T period, Refers to QU i Each user The sum of the number of calls at night during period T; 4) Evening peak call frequency: The evening peak period is defined as 17:00-19:00, and the "evening peak call frequency" indicator is proposed, that is, Refers to the user The call frequency during the evening peak period in period T: in, It means that within the T period, user U i Q with user v Total number of calls during the evening peak period, Refers to the user The total number of calls in period T; For QU i Each user Constructing feature sets Step 5: Use the classification rules of the decision tree to remove users who are unlikely to have a family relationship, based on the number of days in which the nighttime residence is the same in the feature set P. and call days Two indicators are used to judge. When the number of days that two users have the same nighttime residence is 0, it is impossible that the two users have a family relationship in terms of residence. When the number of days that the two users have the same nighttime residence is greater than 0, if the number of days that the two users make calls within a period is greater than 0, it is considered that the two users "may have a family relationship", otherwise it is not a family relationship. Based on this, the feature set of user pairs that "may have a family relationship" is extracted. f is a natural number, and f>0, v is a natural number, and v>0, G f U i Refers to the user Step 6: Family relationship classification based on KNN algorithm: Based on the known family user pairs and non-family user pairs in the artificial samples, construct the user pair feature set W and calibrate the relationship between the user pairs. n is a natural number, and n>0, S a H a Refers to the user L aa Refers to the user The family relationship label between them, the feature set for users who "may have family relationships" f is a natural number, and f>0, v is a natural number, and v>0, and the KNN algorithm is further used for classification. The process is as follows: (1) First, calculate the feature set of users who may have family relationships f is a natural number, and f>0, v is a natural number, and v>0, and the set W of known labels The Euclidean distance between them is calculated as Construct the Euclidean distance set E = {E1, E2, ..., E a }, E a Indicates the user With users The Euclidean distance between (2) Sort the Euclidean distance set E in ascending order according to the distance value; (3) Select the first three samples, determine the frequency of occurrence of the categories to which the first three samples belong, and output the category with the highest frequency; Step 7: Family relationship identification: When the highest frequency category among the three samples is family relationship, the user If the category with the highest frequency in the current three samples is non-family relationship, the user Non-family relationship.
