WO2018120428A1 - 个性化场景预测方法、装置、设备和存储介质 - Google Patents

个性化场景预测方法、装置、设备和存储介质 Download PDF

Info

Publication number
WO2018120428A1
WO2018120428A1 PCT/CN2017/076475 CN2017076475W WO2018120428A1 WO 2018120428 A1 WO2018120428 A1 WO 2018120428A1 CN 2017076475 W CN2017076475 W CN 2017076475W WO 2018120428 A1 WO2018120428 A1 WO 2018120428A1
Authority
WO
WIPO (PCT)
Prior art keywords
scene
user
matrix
prediction
scenario
Prior art date
Application number
PCT/CN2017/076475
Other languages
English (en)
French (fr)
Inventor
刘睿恺
王建明
肖京
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2018120428A1 publication Critical patent/WO2018120428A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02—Services making use of location information
    • H04W4/029—Location-based management or tracking services

Definitions

  • the present invention relates to the field of information processing technologies, and in particular, to a personalized scene prediction method, apparatus, device, and storage medium.
  • the product/service provider needs to predict the target user behavior when providing the product/service to the target user, so that the product/service provider can provide the target user with a product/service that is more in line with their needs, in order to achieve a win-win goal.
  • financial institutions such as banks and insurance provide risk-based financial products to target users
  • the target users are tracked and predicted in real time based on the user behavior scenario prediction method, so that the financial institution predicts the next moment based on the scene of the target user's current time.
  • the scenario has a huge impact on debt recovery or other business promotion.
  • the existing user behavior scene prediction method the amount of user behavior data required to be collected during behavior prediction is large and the value density is low, resulting in a slow behavior prediction process and low accuracy of prediction results.
  • the invention provides a personalized scene prediction method, device, device and storage medium, so as to solve the existing user behavior scene prediction method, the amount of user behavior data to be collected is large and the value density is low, resulting in a slow and predictive behavior prediction process. The result is less accurate.
  • the present invention provides a personalized scene prediction method, including:
  • the geographic location information including POI information associated with the time
  • the present invention provides a personalized scenario prediction apparatus, including:
  • a location information obtaining module configured to acquire geographic location information of the user based on the location service, where the geographic location information includes POI information associated with the time;
  • a trajectory vector sequence acquisition module configured to perform cluster analysis on all geographic location information of the user during a preset period, and obtain a sequence of living habit trajectory vectors
  • a transition matrix building module configured to construct a Markov transition matrix based on the living habit trajectory vector sequence
  • a predicted scene obtaining module configured to acquire a current scene of the user, and obtain a corresponding predicted scene from the Markov transition matrix based on the current scene.
  • the present invention provides a personalized scenario prediction device, including a processor and a memory, the memory storing computer executable instructions for executing the computer executable instructions to perform the following steps:
  • the geographic location information including POI information associated with the time
  • the present invention provides a non-transitory computer readable storage medium for storing one or more computer-executable instructions that are executed by one or more processors such that the one The plurality of processors execute the personalized scene prediction method.
  • the present invention has the following advantages: in the personalized scene prediction method, device, device and storage medium provided by the present invention, clustering analysis is performed on the geographical location information acquired by the user in a preset period.
  • the sequence of life habit trajectory vector because of the strong objectivity and reliability of the geographical location information, the life habit trajectory vector sequence formed by it also has strong objectivity and reliability.
  • the Markov transition matrix is constructed. The data collected by the Markov transition matrix is small, and the calculation process is simple and convenient. Since the Markov transition matrix can clearly show the transition probability from any scene to the next scene, when the predicted scene is acquired based on the Markov transition matrix, the accuracy of the obtained predicted scene is high.
  • FIG. 1 is a flowchart of a personalized scene prediction method in a first embodiment of the present invention
  • FIG. 3 is a schematic block diagram of a personalized scene prediction apparatus in a second embodiment of the present invention.
  • FIG. 4 is a schematic diagram of a personalized scene prediction device in a third embodiment of the present invention.
  • the personalized scenario prediction method may be performed by a personalized scenario prediction device in a financial institution or other product/service provider, for implementing a prediction of a user behavior scenario, so as to facilitate service promotion.
  • the personalized scene prediction method includes the following steps:
  • S10 Acquire location information of the user based on the location service, where the geographic location information includes POI information associated with the time.
  • the geographical location information includes POI information of 0:00-24:00, and each POI information is used to indicate a point in the electronic map, including POI point name, longitude and latitude. And other information. Based on the user's geographic location information, you can find out the home address, office space, shopping places, entertainment places, fitness places, etc. that the user passes every day. It can be understood that obtaining the geographic location information of the user based on the location service has strong objectivity and reliability.
  • LBS Location Based Service
  • GIS Geographic Information System
  • LBS is to obtain the location information (geographic coordinates, or geodetic) of the mobile terminal user through the telecommunication mobile operator's radio communication network (such as GSM network, CDMA network) or external positioning method (such as GPS). Coordinates, a value-added service that provides users with corresponding services, supported by the Geographic Information System (GIS) platform.
  • GIS Geographic Information System
  • LBS is a combination of a mobile communication network and a computer network, and the two networks interact through a gateway.
  • the mobile terminal sends a request through the mobile communication network and transmits it to the LBS service platform through the gateway; the LBS service platform processes according to the user request and the current location of the user, and returns the result to the user through the gateway.
  • POI Point of Interest
  • POI Point of Interest
  • the mobile terminal based on the location service is a smart phone
  • the location function of the smart phone is enabled to enable the LBS service platform to obtain the geographical location information of the smart phone in real time, so as to understand the location of the user carrying the smart phone.
  • Location information The LBS service platform is connected to the personalized scene prediction device of the financial institution or other product/service provider, so that the personalized scenario prediction device can obtain the geographical location information of the user corresponding to the smart phone in real time.
  • the location information includes time and time in the POI information associated with the time, by which the POI information of the user at any moment can be known. It can be understood that the geographical location information is associated with the user ID, and the user ID is used to identify the uniquely identified user, which may be an identity card number or a mobile phone number.
  • the time threshold may be preset, so that when the location service obtains the geographic location information of the user, only the POI information that the user stays at any location reaches the time threshold is obtained. The amount of data of the collected POI information associated with time is avoided, resulting in a problem of low processing efficiency.
  • S20 Perform cluster analysis on all geographic location information of the user during the preset period to obtain a sequence of living habit trajectory vectors.
  • the life habit trajectory vector sequence is composed of trajectory points sorted according to chronological order.
  • the track point is the place where the user passes in daily life, and may be a home address, an office place, a shopping place, an entertainment place, a fitness place, etc., and can be displayed on an electronic map.
  • the preset period may be any period of time before the current system time, and may be one week, one month, three months, or half a year, and may be set independently according to requirements. It can be understood that the longer the preset period, the more the data amount of the geographical location information collected, the higher the accuracy of the processing result; the shorter the preset period, the higher the processing efficiency.
  • the preset period may be set to 1 week to facilitate calculation.
  • step S20 specifically includes the following steps:
  • S21 The DBSCAN algorithm is used to cluster all POI information of any user in a preset period to obtain several sub-clusters.
  • DBSCAN Density-Based Spatial Clustering of Applications with Noise
  • the algorithm divides regions of sufficient density into clusters and finds clusters of arbitrary shape in a spatial database with noise, which defines the cluster as the largest set of points connected by density.
  • the DBSCAN algorithm has the advantages of fast clustering and efficient processing of noise and the discovery of arbitrarily formed spatial clustering.
  • the preset scan radius (hereinafter referred to as eps) and the minimum included point (minPts) in the DBSCAN algorithm are set in advance, and an unvisited POI information is selected to find the distance between them in eps. All POI information (including eps) is output as POI information and all POI information within the eps as a sub-cluster.
  • S22 Perform an iterative aggregation on each sub-cluster by using the K-MEANS algorithm, obtain centroid POI information of each sub-cluster, and output the centroid POI information as a track point.
  • the K-MEANS algorithm is a typical distance-based algorithm.
  • the distance is used as the evaluation index of similarity. That is, the closer the distance between two objects is, the greater the similarity is.
  • Its calculation formula is Among them, the selection of the k initial cluster center points has a great influence on the clustering result, because in the first step of the algorithm, any k objects are randomly selected as the center of the initial cluster, initially representing a cluster. .
  • the algorithm reassigns each object to the nearest cluster for each object remaining in the dataset in each iteration based on its distance from each cluster center. If the value of J does not change before and after an iteration, the algorithm has converged.
  • K-MEANS algorithm can quickly and easily cluster data, has high efficiency and scalability for large data sets, time complexity is nearly linear, and is suitable for mining large-scale data sets.
  • the K-MEANS algorithm is used to iteratively aggregate the POI information in each sub-cluster until the last iteration, and the values before and after the iteration do not change, then the centroid POI information of the sub-cluster is obtained, and the centroid POI information is obtained. Corresponds to a track point.
  • S23 Determine a sequence of lifestyle trajectory vectors of the user during the preset period based on the time sequence of the track points.
  • a sequence of lifestyle trajectory vectors formed by trajectory points sorted in chronological order is obtained.
  • the life habit trajectory vector sequence can clearly reflect the trajectory points of the user's daily home address, office space, shopping place, entertainment place, fitness place, etc., and has strong objectivity and reliability.
  • A is a home address
  • B is an office space
  • C is a shopping place
  • D is an entertainment place
  • E is a fitness place
  • F is a park
  • G is a hospital, etc.
  • A' and A" are Locations within 500m of A, B' and B" are locations within 500m of B, C' and C" are locations within 500m of C, D' and D" are locations within 500m of D, D' and D "A location within 500m near D, D' and D" are locations within 500m of D, ... G' and G" are locations within 500m of G.
  • the geographical location information of the first day includes POI information such as A, A', B', B, C", C, B", B, E", E, A", A; the geographical location information of the next day includes A, A', B', B POI information such as D", D, B", B, F", F, A", A, etc.
  • the scan radius (eps) is 500m and the minimum inclusion point (minPts) is 1, to output A, A', and A" as a sub-cluster, and B, B', B" as a sub-cluster output... G, G ', G' is output as a sub-cluster.
  • step S22 the K-MEANS algorithm is used to cluster each sub-cluster to obtain the centroid POI information in the sub-cluster.
  • the K-MEANS algorithm is used for iterative clustering.
  • the obtained centroid POI information is A, A is output as a track point, and so on, and other track points B, C, D, E, F, and G are acquired.
  • the frequency of the centroid POI information in any sub-cluster is greater than the frequency of other POI information.
  • step S23 based on the time series of the track points, the daily life habit trajectory vector sequence of the user during the preset period is obtained, for example, the track points of the first day are A, B, C, B, E, A, and the next day.
  • the track points are A, B, D, B, F, A, etc.
  • the Markov transition matrix is a stochastic time series model based on probability established by Markov analysis.
  • the obtained Markov transition matrix is as follows:
  • step S30 specifically includes the following steps:
  • each track point in the sequence of lifestyle trajectory vectors corresponds to a scene.
  • all the scenes appearing in the life habit trajectory vector sequence are obtained, that is, all the trajectory points that the user passes during the preset period are counted. If the user's first-day habit trajectory vector sequence is A, B, C, B, E, A; the next day life habit trajectory vector sequence is A, B, D, B, F, A, etc., then the user is two days
  • All scenes appearing in the sequence of lifestyle trajectory vectors include track points (ie, scenes) such as A, B, C, D, E, and F. It can be understood that all scenes appearing in the sequence of lifestyle trajectory vectors can define the size of the finally formed Markov transition matrix, that is, the number of rows and the number of columns defining the Markov transition matrix.
  • the next scenario corresponding to the scenario is obtained from all the life habit trajectory vector sequences in the preset period, and the total number of all the next scenes and the number of occurrences of each next scene are counted to calculate The transition probability of a scene and the next scene is used to construct a Markov transition matrix using the transition probability.
  • the X k+1 scenes for each column at time t k+1, respectively, where the X k+1 scene is the next scene of the X k scene.
  • the transition probabilities of each X k scene to the X k+1 scene are filled in the matrix to construct a Markov transition matrix.
  • the Markov transition matrix can clearly display all the scenes that the user passes during the prediction period, and each scene is acquired based on the user's geographical location information, has objectivity and accuracy, and can clearly display from any scene to the next scene.
  • the probability of transition, the amount of data to be collected is small, and the accuracy of the prediction result is high, which can achieve a more accurate prediction of user behavior, so as to better carry out business promotion.
  • A is the home address
  • C is the office space
  • B, D, E, F, G, H, I, K, and L correspond to other activities except the home address and office space, including but not limited to consumption. (including eating and spending), entertainment, shopping, fitness, etc.
  • all the scenes that the user appears in the life habit trajectory vector sequence within 1 week include 12, such as A, B, C, D, E, F, G, H, I, J, K, and L, so Construct a 12*12 Markov transfer matrix.
  • the Markov transition matrix obtained by calculating the transition probability of each scene and the next scene separately is as follows.
  • the Markov transition matrix can clearly show the transition probability from any scene to the next scene.
  • the amount of data to be collected is small and the accuracy of the prediction result is high, which can accurately predict the user behavior scene, so as to facilitate Good business promotion, etc.
  • each track point can also be associated with the time of the track point, and the Markov transfer matrix can be constructed based on the track points associated with the time of the moment, which can further improve the Mal. Accuracy and reliability of user behavior scenarios in the Cove Transfer Matrix. For example, in the life habit trajectory vector sequence of the user during the preset period, the probability of all the trajectory points and each trajectory point in the same time range (such as 10:00 am) in the preset period is calculated in units of hours, based on The chronological order obtains the conversion probability of any scene and the next scene, and constructs the Markov transition matrix, so that the formed Markov transition matrix is associated with the moment of the trajectory point, further improving the accuracy and reliability of the user behavior scene prediction. Sex.
  • S40 Acquire a current scene of the user, and obtain a corresponding predicted scene from the Markov transition matrix based on the current scene.
  • the Markov transition matrix can clearly show the transition probability from any scene to the next scene.
  • the user's current scene can be obtained, and all possible transitions can be obtained from the Markov transition matrix.
  • a scenario and a transition probability of each next scenario are selected according to the transition probability, and the next scenario with a high transition probability is selected as the predicted scenario, so as to perform a service promotion activity for the user based on the obtained predicted scenario.
  • the clustering analysis is performed on the geographical location information acquired by the user in the preset period to obtain the life habit trajectory vector sequence, because the geographical location information has strong objectivity and Relying on sex, the sequence of life habit trajectory vector formed by it also has strong objectivity and reliability.
  • the Markov transition matrix is constructed. The data collected by the Markov transition matrix is small, and the calculation process is simple and convenient. Since the Markov transition matrix can clearly show the transition probability from any scene to the next scene, when the predicted scene is acquired based on the Markov transition matrix, the accuracy of the obtained predicted scene is high.
  • the personalized scene prediction method further includes the following steps:
  • the normalized transfer matrix is a matrix with high similarity to multiple Markov transfer matrices, and multiple Markov transfer matrices can be converted into a normalized transfer matrix and stored to save storage space.
  • Step S50 specifically includes the following steps:
  • S51 Acquire a Markov transition matrix of multiple users, and each Markov transition matrix is associated with a user ID.
  • the user ID is used to uniquely identify the user, and the user ID is associated with the Markov transition matrix to implement the user corresponding to the Markov transition matrix by the user ID, so as to implement personalized prediction of the user behavior scenario.
  • S52 Perform logistic regression processing on multiple Markov transition matrices to obtain a normalized transfer matrix.
  • Logistic Regression model is used to perform logistic regression processing on multiple Markov transition matrices to obtain a normalized transition matrix.
  • the normalized transition matrix has high similarity with multiple Markov transition matrices. Based on the normalized transfer matrix, the prediction effect of the user behavior scene is similar to that of the corresponding Markov transition matrix, and the normalized conversion matrix can save a lot of storage space.
  • S53 Store the normalized transfer matrix in association with multiple user IDs.
  • the normalized transfer matrix is stored in association with the user ID, that is, the user ID corresponding to the multiple Markov transfer matrices of the normalized transfer matrix is stored in association with the normalized transfer matrix to implement any
  • the user ID can obtain its corresponding normalized transfer matrix, and perform user behavior scenario prediction based on the normalized transfer matrix.
  • the normalized transfer matrix is stored in association with a plurality of user IDs, and the Markov transfer matrix corresponding to the plurality of user IDs is not stored, which can greatly save storage space.
  • the personalized scene prediction method further includes the following steps:
  • S60 Perform scene prediction based on a normalized transfer matrix.
  • the normalized transfer matrix is a matrix with high similarity to multiple Markov transition matrices, based on the normalized transfer matrix for predicting user behavior, its prediction results are predicted by the Markov transition matrix for user behavior scenarios.
  • the prediction results are also highly similar, so that when the normalized transfer matrix is used to predict the user behavior scene, the prediction results also have high accuracy and objectivity.
  • Step S60 specifically includes the following steps:
  • S61 Acquire a scenario prediction request, where the scenario prediction request includes a user ID and a current scenario.
  • the financial institution or other product/service provider may input the user ID corresponding to the user whose behavior prediction is to be performed to the personalized scenario prediction device, and locate the user based on the user ID to determine the corresponding geographic location information. And determining the current scene of the user, so that the personalized scene prediction device acquires the scene prediction request.
  • S62 Determine a normalized transfer matrix of a similar user corresponding to the user ID based on the user ID in the scenario prediction request.
  • the normalized transfer matrix is stored in association with a plurality of user IDs, and the personalized scenario prediction device can query the normalized transfer matrix corresponding to the user ID based on the obtained scenario prediction request, so as to utilize the normalized transfer.
  • the matrix performs scene prediction on the user.
  • S63 Acquire a prediction scenario from the normalized transition matrix based on the current scenario in the scenario prediction request.
  • the normalized transfer matrix is obtained by logistic regression processing by a plurality of users' Markov transition matrices
  • the normalized transition matrix can also clearly show the transition probability of any scene to the next scene, so as to be based on scene prediction.
  • the current scene in the request obtains the transition probability corresponding to the plurality of next scenes and each of the next scenes, so as to output the next scene with a higher transition probability as the predicted scene, so as to improve the accuracy and objectivity of the scene prediction, and It can save storage space.
  • FIG. 3 is a block diagram showing the principle of the personalized scene prediction apparatus in this embodiment.
  • the personalized scenario prediction device may be a personalized scenario prediction device set on a financial institution or other product/service provider, for implementing a prediction of the user behavior scenario, so as to facilitate service promotion.
  • the personalized scene prediction apparatus includes a processor, and the processor is provided with a position information acquisition module 10, a trajectory vector sequence acquisition module 20, a transition matrix construction module 30, a prediction scene acquisition module 40, and a normalization matrix.
  • the function module of the module 50 and the normalized scene prediction module 60 is obtained.
  • the location information obtaining module 10 is configured to acquire geographic location information of the user based on the location service, where the geographic location information includes POI information associated with the time.
  • the geographical location information includes POI information of 0:00-24:00, and each POI information is used to indicate a point in the electronic map, including POI point name, longitude and latitude. And other information. Based on the user's geographic location information, you can find out the home address, office space, shopping places, entertainment places, fitness places, etc. that the user passes every day. It can be understood that obtaining the geographic location information of the user based on the location service has strong objectivity and reliability.
  • LBS Location Based Service
  • GSM Global System for Mobile communications
  • LBS is to obtain the location information of mobile terminal users through the telecommunication mobile operator's radio communication network (such as GSM network, CDMA network) or external positioning mode (such as GPS).
  • GIS Geographic Information System
  • LBS is a combination of a mobile communication network and a computer network, and the two networks interact through a gateway.
  • the mobile terminal sends a request through the mobile communication network and transmits it to the LBS service platform through the gateway; the LBS service platform processes according to the user request and the current location of the user, and returns the result to the user through the gateway.
  • POI Point of Interest
  • POI Point of Interest
  • the mobile terminal based on the location service is a smart phone, and the location function of the smart phone is enabled, so that the LBS service platform obtains the geographical location information of the smart phone in real time, thereby knowing the geographical location information of the user carrying the smart phone.
  • the LBS service platform is connected to the personalized scene prediction device of the financial institution or other product/service provider, so that the personalized scenario prediction device can obtain the geographical location information of the user corresponding to the smart phone in real time.
  • the location information includes time and time in the POI information associated with the time, by which the POI information of the user at any moment can be known. It can be understood that the geographical location information is associated with the user ID, and the user ID is used to identify the uniquely identified user, which may be an identity card number or a mobile phone number.
  • the time threshold may be preset, so that when the location service obtains the geographic location information of the user, only the POI information that the user stays at any location reaches the time threshold is obtained. The amount of data of the collected POI information associated with time is avoided, resulting in a problem of low processing efficiency.
  • the trajectory vector sequence obtaining module 20 is configured to perform cluster analysis on all geographic location information of the user during the preset period to obtain a living habit trajectory vector sequence.
  • the life habit trajectory vector sequence is composed of trajectory points sorted according to chronological order.
  • the track point is the place where the user passes in daily life, and may be a home address, an office place, a shopping place, an entertainment place, a fitness place, etc., and can be displayed on an electronic map.
  • the preset period may be any period of time before the current system time, and may be one week, one month, three months, or half a year, and may be set independently according to requirements. It can be understood that the longer the preset period, the more the data amount of the geographical location information collected, the higher the accuracy of the processing result; the shorter the preset period, the higher the processing efficiency.
  • the preset period may be set to 1 week to facilitate calculation.
  • trajectory vector sequence acquisition module 20 specifically includes a sub-cluster acquisition unit 21, a trajectory point acquisition unit 22, and a vector sequence acquisition unit 23.
  • the sub-cluster acquiring unit 21 is configured to use the DBSCAN algorithm to cluster all POI information of any user in a preset period to obtain several sub-clusters.
  • DBSCAN Density-Based Spatial Clustering of Applications with Noise
  • the algorithm divides regions of sufficient density into clusters and finds clusters of arbitrary shape in a spatial database with noise, which defines the cluster as the largest set of points connected by density.
  • the DBSCAN algorithm has the advantages of fast clustering and efficient processing of noise and the discovery of arbitrarily formed spatial clustering.
  • the preset scan radius (hereinafter referred to as eps) and the minimum included point (minPts) in the DBSCAN algorithm are set in advance, and an unvisited POI information is selected to find the distance between them in eps. All POI information (including eps) is output as POI information and all POI information within the eps as a sub-cluster.
  • the track point obtaining unit 22 is configured to perform iterative aggregation on each sub-cluster by using the K-MEANS algorithm, acquire centroid POI information of each sub-cluster, and output the centroid POI information as a track point.
  • the K-MEANS algorithm is a typical distance-based algorithm.
  • the distance is used as the evaluation index of similarity. That is, the closer the distance between two objects is, the greater the similarity is.
  • Its calculation formula is Among them, the selection of the k initial cluster center points has a great influence on the clustering result, because in the first step of the algorithm, any k objects are randomly selected as the center of the initial cluster, initially representing a cluster. .
  • the algorithm reassigns each object to the nearest cluster for each object remaining in the dataset in each iteration based on its distance from each cluster center. If the value of J does not change before and after an iteration, the algorithm has converged.
  • K-MEANS algorithm can quickly and easily cluster data, has high efficiency and scalability for large data sets, time complexity is nearly linear, and is suitable for mining large-scale data sets.
  • the K-MEANS algorithm is used to iteratively aggregate the POI information in each sub-cluster until the last iteration, and the values before and after the iteration do not change, then the centroid POI information of the sub-cluster is obtained, and the centroid POI information is obtained. Corresponds to a track point.
  • the vector sequence obtaining unit 23 is configured to determine a sequence of lifestyle trajectory vectors of the user during the preset period based on the time series of the track points.
  • a sequence of lifestyle trajectory vectors formed by trajectory points sorted in chronological order is obtained.
  • the life habit trajectory vector sequence can clearly reflect the trajectory points of the user's daily home address, office space, shopping place, entertainment place, fitness place, etc., and has strong objectivity and reliability.
  • A is a home address
  • B is an office space
  • C is a shopping place
  • D is an entertainment place
  • E is a fitness place
  • F is a park
  • G is a hospital, etc.
  • A' and A" are Location within 500m of A
  • B' and B" For locations within 500m of B, C' and C" are locations within 500m of C, D' and D" are locations within 500m of D, D' and D" are locations within 500m of D, D' and D" is a location within 500m near D, ... G' and G" are locations within 500m of G.
  • the geographic location information of the first day includes A, A', B', B, C" POI information such as C, B", B, E", E, A", A; the geographical location information of the next day includes A, A', B', B, D", D, B", B, F ", F, A", A, etc. POI information, etc.
  • all POI information in one week is set to a scanning radius (eps) of 500 m and a minimum inclusion point ( minPts) is 1, to output A, A', A" as a sub-cluster, and B, B', B" as a sub-cluster output... G, G', G" are output as a sub-cluster.
  • step S22 the K-MEANS algorithm is used to cluster each sub-cluster to obtain the centroid POI information in the sub-cluster.
  • the K-MEANS algorithm is used for iterative clustering.
  • the obtained centroid POI information is A, A is output as a track point, and so on, and other track points B, C, D, E, F, and G are acquired.
  • the centroid POI information in any sub-cluster The frequency of occurrence is greater than the frequency of occurrence of other POI information.
  • step S23 based on the time sequence of the track point, the daily life habit trajectory vector sequence of the user during the preset period is obtained, for example, the track point of the first day is A, B, C, B, E, A, the track points of the next day are A, B, D, B, F, A, etc.
  • the transfer matrix construction module 30 is configured to construct a Markov transfer matrix based on the life habit trajectory vector sequence.
  • the Markov transition matrix is a stochastic time series model based on probability established by Markov analysis.
  • the obtained Markov transition matrix is as follows:
  • the transition matrix construction module 30 specifically includes a scene acquisition unit 31, a probability calculation unit 32, and a matrix construction unit 33.
  • the scene obtaining unit 31 is configured to acquire all scenes appearing in the sequence of the lifestyle trajectory vector based on the life habit trajectory vector sequence.
  • each track point in the sequence of lifestyle trajectory vectors corresponds to a scene.
  • all the scenes appearing in the life habit trajectory vector sequence are obtained, that is, all the trajectory points that the user passes during the preset period are counted. If the user's first-day habit trajectory vector sequence is A, B, C, B, E, A; the next day life habit trajectory vector sequence is A, B, D, B, F, A, etc., then the user is two days
  • All scenes appearing in the sequence of lifestyle trajectory vectors include track points (ie, scenes) such as A, B, C, D, E, and F. It can be understood that all scenes appearing in the sequence of lifestyle trajectory vectors can define the size of the finally formed Markov transition matrix, that is, the number of rows and the number of columns defining the Markov transition matrix.
  • the probability calculation unit 32 is configured to calculate a transition probability of any scene and the next scene.
  • the next scenario corresponding to the scenario is obtained from all the life habit trajectory vector sequences in the preset period, and the total number of all the next scenes and the number of occurrences of each next scene are counted to calculate The transition probability of a scene and the next scene is used to construct a Markov transition matrix using the transition probability.
  • the matrix construction unit 33 is configured to construct a Markov transfer matrix based on the transition probability.
  • the X k+1 scenes for each column at time t k+1, respectively, where the X k+1 scene is the next scene of the X k scene.
  • the transition probabilities of each X k scene to the X k+1 scene are filled in the matrix to construct a Markov transition matrix.
  • the Markov transition matrix can clearly display all the scenes that the user passes during the prediction period, and each scene is acquired based on the user's geographical location information, has objectivity and accuracy, and can clearly display from any scene to the next scene.
  • the probability of transition, the amount of data to be collected is small, and the accuracy of the prediction result is high, which can achieve a more accurate prediction of user behavior, so as to better carry out business promotion.
  • A is the home address
  • C is the office space
  • B, D, E, F, G, H, I, K, and L correspond to other activities except the home address and office space, including but not limited to consumption. (including eating and spending), entertainment, shopping, Fitness and more.
  • all the scenes that the user appears in the life habit trajectory vector sequence within 1 week include 12, such as A, B, C, D, E, F, G, H, I, J, K, and L, so Construct a 12*12 Markov transfer matrix.
  • the Markov transition matrix obtained by calculating the transition probability of each scene and the next scene separately is as follows.
  • the Markov transition matrix can clearly show the transition probability from any scene to the next scene.
  • the amount of data to be collected is small and the accuracy of the prediction result is high, which can accurately predict the user behavior scene, so as to facilitate Good business promotion, etc.
  • each track point can also be associated with the time of the track point, and the Markov transfer matrix can be constructed based on the track points associated with the time of the moment, which can further improve the Mal. Accuracy and reliability of user behavior scenarios in the Cove Transfer Matrix. For example, in the life habit trajectory vector sequence of the user during the preset period, the probability of all the trajectory points and each trajectory point in the same time range (such as 10:00 am) in the preset period is calculated in units of hours, based on The chronological order obtains the conversion probability of any scene and the next scene, and constructs the Markov transition matrix, so that the formed Markov transition matrix is associated with the moment of the trajectory point, further improving the accuracy and reliability of the user behavior scene prediction. Sex.
  • the predicted scene obtaining module 40 is configured to acquire a current scene of the user, and obtain a corresponding predicted scene from the Markov transition matrix based on the current scene.
  • the Markov transition matrix can clearly show the transition probability from any scene to the next scene.
  • the user's current scene can be obtained, and all possible transitions can be obtained from the Markov transition matrix.
  • the next scenario with a higher transition probability is selected as a pre- The scenario is measured to conduct a business promotion activity for the user based on the obtained predicted scenario.
  • the clustering analysis is performed on the geographical location information acquired by the user in the preset period to obtain the life habit trajectory vector sequence, because the geographical location information has strong objectivity and reliability.
  • sexuality the sequence of life habit trajectory vector formed by it also has strong objectivity and reliability.
  • the Markov transition matrix is constructed. The data collected by the Markov transition matrix is small, and the calculation process is simple and convenient. Since the Markov transition matrix can clearly show the transition probability from any scene to the next scene, when the predicted scene is acquired based on the Markov transition matrix, the accuracy of the obtained predicted scene is high.
  • the personalized scene prediction apparatus further includes a normalized matrix acquisition module 50 for acquiring a normalized transfer matrix based on the Markov transition matrix.
  • the normalized transfer matrix is a matrix with high similarity to multiple Markov transfer matrices, and multiple Markov transfer matrices can be converted into a normalized transfer matrix and stored to save storage space.
  • the normalized matrix acquisition module 50 specifically includes a matrix acquisition unit 51, a logical regression processing unit 52, and a matrix association storage unit 53.
  • the matrix obtaining unit 51 is configured to acquire a Markov transition matrix of a plurality of users, and each Markov transition matrix is associated with a user ID.
  • the user ID is used to uniquely identify the user, and the user ID is associated with the Markov transition matrix to implement the user corresponding to the Markov transition matrix by the user ID, so as to implement personalized prediction of the user behavior scenario.
  • the logistic regression processing unit 52 is configured to perform a logistic regression process on the plurality of Markov transition matrices to obtain a normalized transfer matrix.
  • Logistic Regression model is used to perform logistic regression processing on multiple Markov transition matrices to obtain a normalized transition matrix.
  • the normalized transition matrix has high similarity with multiple Markov transition matrices. Based on the normalized transfer matrix, the prediction effect of the user behavior scene is similar to that of the corresponding Markov transition matrix, and the normalized conversion matrix can save a lot of storage space.
  • the matrix association storage unit 53 is configured to store the normalized transfer matrix in association with a plurality of user IDs.
  • the normalized transfer matrix is stored in association with the user ID, that is, the user ID corresponding to the multiple Markov transfer matrices of the normalized transfer matrix is stored in association with the normalized transfer matrix to implement any
  • the user ID can obtain its corresponding normalized transfer matrix, and perform user behavior scenario prediction based on the normalized transfer matrix.
  • the normalized transfer matrix is stored in association with a plurality of user IDs, and the Markov transfer matrix corresponding to the plurality of user IDs is not stored, which can greatly save storage space.
  • the personalized scene prediction apparatus further includes a normalized scene prediction module 60 for The scene is predicted by the normalized transfer matrix.
  • the normalized transfer matrix is a matrix with high similarity to multiple Markov transition matrices, based on the normalized transfer matrix for predicting user behavior, its prediction results are predicted by the Markov transition matrix for user behavior scenarios.
  • the prediction results are also highly similar, so that when the normalized transfer matrix is used to predict the user behavior scene, the prediction results also have high accuracy and objectivity.
  • the normalized scene prediction module 60 specifically includes a prediction request acquisition unit 61, a normalization matrix acquisition unit 62, and a normalized scene prediction unit 63.
  • the prediction request obtaining unit 61 is configured to acquire a scene prediction request, where the scene prediction request includes a user ID and a current scene.
  • the financial institution or other product/service provider may input the user ID corresponding to the user whose behavior prediction is to be performed to the personalized scenario prediction device, and locate the user based on the user ID to determine the corresponding geographic location information. And determining the current scene of the user, so that the personalized scene prediction device acquires the scene prediction request.
  • the normalization matrix obtaining unit 62 is configured to determine a normalized transfer matrix of a similar user corresponding to the user ID based on the user ID in the scenario prediction request.
  • the normalized transfer matrix is stored in association with a plurality of user IDs, and the personalized scenario prediction device can query the normalized transfer matrix corresponding to the user ID based on the obtained scenario prediction request, so as to utilize the normalized transfer.
  • the matrix performs scene prediction on the user.
  • the normalized scenario prediction unit 63 is configured to obtain a predicted scenario from the normalized transition matrix based on the current scenario in the scenario prediction request.
  • the normalized transfer matrix is obtained by logistic regression processing by a plurality of users' Markov transition matrices
  • the normalized transition matrix can also clearly show the transition probability of any scene to the next scene, so as to be based on scene prediction.
  • the current scene in the request obtains the transition probability corresponding to the plurality of next scenes and each of the next scenes, so as to output the next scene with a higher transition probability as the predicted scene, so as to improve the accuracy and objectivity of the scene prediction, and It can save storage space.
  • FIG. 3 is a schematic structural diagram of a personalized scenario prediction apparatus 300 according to a third embodiment of the present invention.
  • the device 300 may be a mobile terminal, a desktop computer, a server, or the like, such as a mobile phone, a tablet computer, a personal digital assistant (PDA), or an in-vehicle computer having certain data processing capabilities.
  • the device 300 includes a radio frequency (RF) circuit 301, a memory 302, an input module 303, a display module 304, a processor 305, an audio circuit 306, a WiFi (Wireless Fidelity) module 307, and a power source 308.
  • RF radio frequency
  • the input module 303 and the display module 304 are used as user interaction devices of the device 300 for implementing interaction between the user and the device 300, for example, receiving a scene prediction instruction input by the user and displaying a corresponding predicted scene.
  • the input module 303 is configured to receive a scenario prediction instruction input by the user, and send the scenario prediction instruction to the processor 305, where the scenario prediction instruction includes the current scenario.
  • the processor 305 is configured to receive the scenario prediction instruction, and acquire the predicted scenario based on the scenario prediction instruction, and send the predicted scenario to the display module 304.
  • Display module 304 receives and displays the predicted scene.
  • the input module 303 can be configured to receive numeric or character information input by a user, and to generate signal inputs related to user settings and function control of the device 300.
  • the input module 303 can include a touch panel 3031.
  • the touch panel 3031 also referred to as a touch screen, can collect touch operations on or near the user (such as the operation of the user using any suitable object or accessory such as a finger or a stylus on the touch panel 3031), and according to the preset The programmed program drives the corresponding connection device.
  • the touch panel 3031 may include two parts of a touch detection device and a touch controller.
  • the touch detection device detects the touch orientation of the user, and detects a signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts the touch information into contact coordinates, and sends the touch information.
  • the processor 305 is provided and can receive commands from the processor 305 and execute them.
  • the touch panel 3031 can be implemented in various types such as resistive, capacitive, infrared, and surface acoustic waves.
  • the input module 303 may further include other input devices 3032.
  • the other input devices 3032 may include but are not limited to physical keyboards, function keys (such as volume control buttons, switch buttons, etc.), trackballs, mice, joysticks, and the like. One or more of them.
  • display module 304 can be used to display information entered by a user or information provided to a user and various menu interfaces of device 300.
  • the display module 304 can include a display panel 3041.
  • the display panel 3041 can be configured in the form of an LCD or an Organic Light-Emitting Diode (OLED).
  • the touch panel 3031 can cover the display panel 3041 to form a touch display screen.
  • the touch display screen detects a touch operation on or near it, it is transmitted to the processor 305 to determine the type of the touch event, and then processed.
  • the 305 provides a corresponding visual output on the touch display based on the type of touch event.
  • the touch display includes an application interface display area and a common control display area.
  • the arrangement manner of the application interface display area and the display area of the common control is not limited, and the arrangement manner of the two display areas can be distinguished by up-and-down arrangement, left-right arrangement, and the like.
  • the application interface display area can be used to display the interface of the application. Each interface can contain interface elements such as at least one application's icon and/or widget desktop control.
  • the application interface display area can also be an empty interface that does not contain any content.
  • the common control display area is used to display controls with high usage, such as setting buttons, interface numbers, scroll bars, phone book icons, and the like.
  • the WiFi module 307 functions as a network interface of the device 300, and can implement data interaction between the device 300 and other devices.
  • the network interface can be connected to the remote storage device and the external display device through network communication.
  • the network interface is configured to receive the geographic location information and the current scenario sent by the remote storage device, and send the geographic location information and the current scenario to the processor 305;
  • the predicted scene sent by the processor 305 is sent to the external display device.
  • the external display device can receive and display the predicted scene.
  • the remote storage device connected to the network interface through the WiFi network may be a cloud server or other database, where the geographical location information and the current scene are stored on the remote storage device, and the geographic location may be
  • the location information and the current scenario may be sent to the WiFi module 307 through the WiFi network, and the WiFi module 307 sends the acquired geographical location fence information and the current scenario to the processor 305, and sends the predicted scenario.
  • the WiFi module 307 sends the acquired geographical location fence information and the current scenario to the processor 305, and sends the predicted scenario.
  • the external display device may be a cloud server or other database, where the geographical location information and the current scene are stored on the remote storage device, and the geographic location may be
  • the location information and the current scenario may be sent to the WiFi module 307 through the WiFi network, and the WiFi module 307 sends the acquired geographical location fence information and the current scenario to the processor 305, and sends the predicted scenario.
  • the WiFi module 307 sends the acquired geographical location fence information and the current scenario to the processor 305, and sends the
  • the memory 302 includes a first memory 3021 and a second memory 3022.
  • the first memory 3021 can be a non-transitory computer readable storage medium having an operating system, a database, and computer executable instructions stored thereon.
  • Computer executable instructions are executable by processor 305 for implementing a personalized scene prediction method of the embodiment as shown in FIG.
  • the database on the memory 302 is used to store various types of data, such as various data involved in the above-described personalized scene prediction method, such as geographic location information and the Markov transfer matrix.
  • the second memory 3021 can be an internal memory of the device 300 that provides a cached operating environment for operating systems, databases, and computer executable instructions in a non-transitory computer readable storage medium.
  • processor 305 is the control center of device 300, which connects various portions of the entire handset using various interfaces and lines, by running or executing computer-executable collections and/or databases stored in first memory 3021. The data, performing various functions and processing data of the device 300, thereby performing overall monitoring of the device 300.
  • processor 305 can include one or more processing modules.
  • the processor 305 by executing the stored in the computer executable instructions and/or the data in the database in the first memory 3021, the processor 305 is configured to perform the following steps: acquiring geographic location information of the user based on the location service, The geographic location information includes POI information associated with time; clustering analysis of all geographic location information of the user during the preset period to obtain a sequence of living habit trajectory vectors; constructing Markov based on the sequence of the living habit trajectory vector Transferring a matrix; acquiring a current scene of the user, and acquiring a corresponding predicted scene from the Markov transition matrix based on the current scene.
  • the clustering analysis is performed on all geographic location information of the user during the preset period to obtain a sequence of living habit trajectory vectors, including:
  • the DBSCAN algorithm is used to cluster all POI information of any user in a preset period to obtain several sub-clusters;
  • a sequence of lifestyle trajectory vectors of the user during the preset period is determined based on a time series of the track points.
  • the constructing a Markov transition matrix based on the sequence of the living habit trajectory vector comprises:
  • the Markov transition matrix is constructed based on the transition probability.
  • the processor 305 further performs the following steps: acquiring a normalized transfer matrix based on the Markov transfer matrix;
  • the acquiring the normalized transfer matrix based on the Markov transfer matrix includes:
  • the normalized transfer matrix is stored in association with a plurality of the user IDs.
  • the processor 305 further performs the following steps: performing scenario prediction based on the normalized transition matrix;
  • the performing scene prediction based on the normalized transfer matrix includes:
  • the scenario prediction request includes a user ID and a current scenario
  • the processor 305 may acquire geographic location information of the user based on the location service, where the geographic location information includes POI information associated with the time; and aggregate all geographic location information of the user during the preset period.
  • Class analysis obtaining a life habit trajectory vector sequence; constructing a Markov transition matrix based on the living habit trajectory vector sequence; acquiring a current scene of the user, and obtaining corresponding corresponding from the Markov transition matrix based on the current scene
  • the scenario is predicted such that the device 300 can predict the user scenario, and the obtained predicted scenario has high accuracy, objectivity and reliability, and the calculation process is simple and convenient.
  • the life experience custom track vector sequence obtained by the processor 305 performing cluster analysis on the geographic location information may be stored in the processor cache or may be stored in the memory 302; the processor 305 may obtain the predicted scene based on the current scene. It is sent to a user interaction device (such as display unit 304) or a network interface (such as WiFi module 307). It can be understood that the display unit 304 can display the predicted scene; the WiFi module 307 can send the predicted scene to an external display device, and the predicted scene is displayed by the external display device.
  • the embodiment provides a non-transitory computer readable storage medium.
  • the non-transitory computer readable storage medium For storing one or more computer executable instructions.
  • the computer executable instructions are executed by one or more processors, such that the one or more processors perform the personalized scene prediction method described in the first embodiment. To avoid repetition, details are not described herein again.
  • modules and algorithm steps of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the solution. A person skilled in the art can use different methods for implementing the described functions for each particular application, but such implementation should not be considered to be beyond the scope of the present invention.
  • the disclosed apparatus and method may be implemented in other manners.
  • the device embodiments described above are merely illustrative.
  • the division of the modules is only a logical function division.
  • there may be another division manner for example, multiple modules or components may be combined or Can be integrated into another system, or some features can be ignored or not executed.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or module, and may be electrical, mechanical or otherwise.
  • the modules described as separate components may or may not be physically separated.
  • the components displayed as modules may or may not be physical modules, that is, may be located in one place, or may be distributed to multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the embodiment.
  • each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.
  • the functions, if implemented in the form of software functional modules and sold or used as separate products, may be stored in a computer readable storage medium.
  • the technical solution of the present invention which is essential or contributes to the prior art, or a part of the technical solution, may be embodied in the form of a software product, which is stored in a storage medium, including
  • the instructions are used to cause a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention.
  • the foregoing storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a ROM, a RAM, a magnetic disk, or an optical disk.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Quality & Reliability (AREA)
  • Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Development Economics (AREA)
  • Tourism & Hospitality (AREA)
  • Game Theory and Decision Science (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种个性化场景预测方法、装置、设备和存储介质。该个性化场景预测方法包括：基于位置服务获取用户的地理位置信息，所述地理位置信息包括与时间相关联的POI信息（S10）；对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列（S20）；基于所述生活习惯轨迹向量序列，构建马尔科夫转移矩阵（S30）；获取用户的当前场景，基于所述当前场景从所述马尔科夫转移矩阵中，获取对应的预测场景（S40）。该个性化场景预测方法进行用户行为场景预测时，所需采集的数据量少，计算过程简单方便且预测准确性较高。

Description

个性化场景预测方法、装置、设备和存储介质 技术领域
本发明涉及信息处理技术领域，尤其涉及一种个性化场景预测方法、装置、设备和存储介质。
背景技术
随着互联网的发展，人们的生活越来越多地与互联网紧密联系在一起。随着人们日常生活节奏越来越快，人们日益希望通过互联网快速找到自己所需要的产品/服务，以达到节省时间的效果。相应地，产品/服务提供者在给目标用户提供产品/服务时，需对目标用户行为进行预测，以使产品/服务提供者给目标用户提供更符合其需求的产品/服务，以达到双赢目的。如在银行、保险等金融机构给目标用户提供风险类金融产品时，基于用户行为场景预测方法对目标用户进行实时跟踪并预测，以使金融机构基于目标用户当前时间所处场景预测下一时刻所处场景，在债务追讨或者其他业务推广中产生巨大作用。现有用户行为场景预测方法中，在行为预测时所需采集的用户行为数据量较大且价值密度低，导致行为预测过程效率慢且预测结果准确性较低。
发明内容
本发明提供一种个性化场景预测方法、装置、设备和存储介质，以解决现有用户行为场景预测方法时需采集的用户行为数据量较大且价值密度低，导致行为预测过程效率慢且预测结果准确性较低的问题。
本发明解决其技术问题所采用的技术方案是：
第一方面，本发明提供一种个性化场景预测方法，包括：
基于位置服务获取用户的地理位置信息，所述地理位置信息包括与时间相关联的POI信息；
对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列；
基于所述生活习惯轨迹向量序列，构建马尔科夫转移矩阵；
获取用户的当前场景，基于所述当前场景从所述马尔科夫转移矩阵中，获取对应的预 测场景。
第二方面，本发明提供一种个性化场景预测装置，包括：
位置信息获取模块，用于基于位置服务获取用户的地理位置信息，所述地理位置信息包括与时间相关联的POI信息；
轨迹向量序列获取模块，用于对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列；
转移矩阵构建模块，用于基于所述生活习惯轨迹向量序列，构建马尔科夫转移矩阵；
预测场景获取模块，用于获取用户的当前场景，基于所述当前场景从所述马尔科夫转移矩阵中，获取对应的预测场景。
第三方面，本发明提供一种个性化场景预测设备，包括处理器及存储器，所述存储器存储有计算机可执行指令，所述处理器用于执行所述计算机可执行指令以执行如下步骤：
基于位置服务获取用户的地理位置信息，所述地理位置信息包括与时间相关联的POI信息；
对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列；
基于所述生活习惯轨迹向量序列，构建马尔科夫转移矩阵；
获取用户的当前场景，基于所述当前场景从所述马尔科夫转移矩阵中，获取对应的预测场景。
第四方面，本发明提供一种非易失性计算机可读存储介质，用于存储一个或多个计算机可执行指令，所述计算机可执行指令被一个或多个处理器执行，使得所述一个或多个处理器执行所述个性化场景预测方法。
本发明与现有技术相比具有如下优点：本发明所提供的个性化场景预测方法、装置、设备和存储介质中，通过对用户在预设期间内获取的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列，由于地理位置信息具有较强的客观性和可靠性，使其形成的生活习惯轨迹向量序列也具有较强的客观性和可靠性。基于生活习惯轨迹向量序列，构建马尔科夫转移矩阵，马尔科夫转移矩阵构建过程所需采集的数据量少，计算过程简单方便。由于马尔科夫转移矩阵可清楚显示从任一场景到下一场景的转移概率，使得基于马尔科夫转移矩阵获取预测场景时，所获取到的预测场景的准确性较高。
附图说明
下面将结合附图及实施例对本发明作进一步说明，附图中：
图1是本发明第一实施例中个性化场景预测方法的一流程图；
图2是本发明第一实施例中个性化场景预测方法的另一流程图；
图3是本发明第二实施例中个性化场景预测装置的一原理框图。
图4是本发明第三实施例中个性化场景预测设备的一示意图。
具体实施方式
为了对本发明的技术特征、目的和效果有更加清楚的理解，现对照附图详细说明本发明的具体实施方式。
第一实施例
图1和图2示出本实施例中个性化场景预测方法的流程图。该个性化场景预测方法可由金融机构或者其他产品/服务提供者中的个性化场景预测设备执行，用于实现对用户行为场景预测，以便于进行业务推广。如图1和图2所示，该个性化场景预测方法，包括如下步骤：
S10：基于位置服务获取用户的地理位置信息，地理位置信息包括与时间相关联的POI信息。
以任一用户一天的地理位置信息为例，该地理位置信息中包括0：00—24：00的POI信息，每一POI信息用于指示电子地图中的一点，包括POI点名称、经度和纬度等信息。基于用户的地理位置信息，可了解用户每天经过的家庭住址、办公场所、购物场所、娱乐场所、健身场所等信息。可以理解地，基于位置服务获取用户的地理位置信息，具有较强的客观性和可靠性。
基于位置服务(Location Based Service，简称LBS)是通过电信移动运营商的无线电通讯网络(如GSM网、CDMA网)或外部定位方式(如GPS)获取移动终端用户的位置信息(地理坐标，或大地坐标)，在地理信息系统(Geographic Information System，简称GIS)平台的支持下，为用户提供相应服务的一种增值业务。总体来看，LBS由移动通信网络和计算机网络结合而成，两个网络之间通过网关实现交互。移动终端通过移动通信网络发出请求，经过网关传递给LBS服务平台；LBS服务平台根据用户请求和用户当前位置进行处理，并将结果通过网关返回给用户。POI(Point Of Interest,即兴趣点或信息点)，包括名称、类型、经度、纬度等资料，以使POI可在电子地图上呈现，以标示电子地图上的某个地点信息。
本实施例中，基于位置服务的移动终端为智能手机，通过开启智能手机上的定位功能，以使LBS服务平台实时获取智能手机的地理位置信息，从而了解携带该智能手机的用户的地 理位置信息。该LBS服务平台与金融机构或者其他产品/服务提供者中的个性化场景预测设备相连，以使该个性化场景预测设备能够实时获取该智能手机对应的用户的地理位置信息。地理位置信息包括与时间相关联的POI信息中的时间包括日期和时刻，通过该地理位置信息可了解用户在任一时刻所处的POI信息。可以理解地，地理位置信息与用户ID相关联，用户ID用于识别唯一识别用户，可以是身份证号或手机号。
可以理解地，为了减少数据处理量，提高处理效率，可预先设置时间阈值，以使基于位置服务获取用户的地理位置信息时，只获取用户在任一地点停留时间达到该时间阈值的POI信息，以避免采集到的与时间相关联的POI信息的数据量较多，导致处理效率低的问题。
S20：对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列。
其中，生活习惯轨迹向量序列由依据时间顺序排序的轨迹点组成。轨迹点是用户日常生活中经过的地点，可以是家庭住址、办公场所、购物场所、娱乐场所、健身场所等地点，可在电子地图中显示。其中，预设期间可以是当前系统时间之前的任意一段时间，可以为一周、一个月、三个月或半年，可根据需求自主设置。可以理解地，预设期间越长，其采集到的地理位置信息的数据量越多，处理结果的准确性越高；预设期间越短，其处理效率越高。为说明本实施例所提供的个性化场景预设方法的实现用户行为场景预测过程，可将预设期间设为1周，以便于计算。
进一步地，步骤S20具体包括如下步骤：
S21：采用DBSCAN算法对任一用户在预设期间内所有POI信息进行聚类，以获取若干子集群。
其中，DBSCAN(Density-Based Spatial Clustering of Applications with Noise，具有噪声的基于密度的聚类方法)是一种基于密度的空间算法。该算法将具有足够密度的区域划分为簇，并在具有噪声的空间数据库中发现任意形状的簇，它将簇定义为密度相连的点的最大集合。DBSCAN算法具有聚类速度快且能够有效处理噪声和发现任意形成的空间聚类的优点。
本实施例中，预先设置DBSCAN算法中的预设扫描半径(以下简称为eps)和最小包含点数(minPts)，任选一个未被访问(unvisited)的POI信息开始，找出与其距离在eps之内(包括eps)的所有POI信息，将POI信息与距离在eps之内的所有POI信息作为一个子集群输出。
S22：采用K-MEANS算法对每一子集群进行迭代聚合，获取每一子集群的质心POI信息，并将质心POI信息作为轨迹点输出。
K-MEANS算法是很典型的基于距离的算法，采用距离作为相似性的评价指标，即认为两个对象的距离越近，其相似度就越大。其计算公式为
Figure PCTCN2017076475-appb-000001
其中，k个初始类聚类中心点的选取对聚类结果具有较大的影响，因为在该算法第一步中是随机的选取任意k个对象作为初始聚类的中心，初始地代表一个簇。该算法在每次迭代中对数据集中剩余的每个对象，根据其与各个簇中心的距离将每个对象重新赋给最近的簇。若一次迭代前后，J的值没有发生变化，说明算法已经收敛。K-MEANS算法可快速简单地对数据进行聚类，对大数据集具有较高的效率且可伸缩性，时间复杂度近于线性，而且适合挖掘大规模数据集。
本实施例中，采用K-MEANS算法对每一子集群中的POI信息进行迭代聚合，直到最后一次迭代时，迭代前后数值没有发生变化，则获取该子集群的质心POI信息，该质心POI信息对应一轨迹点。
S23：基于轨迹点的时间序列，确定用户在预设期间内的生活习惯轨迹向量序列。
本实施例中，通过对用户在预设期间内每日采集到的地理位置信息进行聚类分析，获取每日由按时间顺序排序的轨迹点形成的生活习惯轨迹向量序列。该生活习惯轨迹向量序列可清楚体现用户每日经过的家庭住址、办公场所、购物场所、娱乐场所、健身场所等轨迹点，具有较强的客观性和可靠性。
在一具体实施方式中，若A为家庭住址，B为办公场所，C为购物场所，D为娱乐场所，E为健身场所，F为公园，G为医院等等；且A’和A”为A附近500m内的地点，B’和B”为B附近500m内的地点，C’和C”为C附近500m内的地点，D’和D”为D附近500m内的地点，D’和D”为D附近500m内的地点，D’和D”为D附近500m内的地点，……G’和G”为G附近500m内的地点。在1周内，第一天的地理位置信息包括A、A’、B’、B、C”、C、B”、B、E”、E、A”、A等POI信息；第二天的地理位置信息包括A、A’、B’、B、D”、D、B”、B、F”、F、A”、A等POI信息……依此类推。步骤S21中采用DBSCAN算法进行聚类时，将1周内所有POI信息，通过设置扫描半径(eps)为500m和最小包含点数(minPts)为1，以将A、A’、A”作为一子集群输出，将B，B’、B”作为一子集群输出……G，G’、G”作为一子集群输出。步骤S22中采用K-MEANS算法对每一子集群进行聚类，获取到子集群中的质心POI信息，对于子集群A、A’、A”而言，采用K-MEANS算法进行迭代聚类时，获取到的质心POI信息为A，将A作为轨迹点输出，依此类推，获取其他轨迹点B、C、D、E、F和G。本实施 例中，任一子集群中质心POI信息出现的频率大于其他POI信息出现的频率。步骤S23中，基于轨迹点的时间序列，获取用户在预设期间内每日的生活习惯轨迹向量序列，如第一天的轨迹点为A、B、C、B、E、A，第二天的轨迹点为A、B、D、B、F、A……等。
S30：基于生活习惯轨迹向量序列，构建马尔科夫转移矩阵。
马尔科夫转移矩阵，即马尔科夫(Markov Process)链的转移概率(transition probability)矩阵，是一种利用马尔科夫分析方法基于概率建立的随机型的时序模型。马尔科夫分析方法的基本模型为：X(k+1)＝X(k)*P，其中，X(k)为预测用户在t＝k时刻的场景向量，P表示一步转移概率矩阵；X(k+1)为预测用户在t＝k+1时刻的场景向量。本实施例中，所获取的马尔科夫转移矩阵如下：
Figure PCTCN2017076475-appb-000002
进一步地，步骤S30具体包括如下步骤：
S31：基于生活习惯轨迹向量序列，获取生活习惯轨迹向量序列中出现的所有场景。
本实施例中，生活习惯轨迹向量序列中的每一轨迹点对应一场景。基于生活习惯轨迹向量序列，获取生活习惯轨迹向量序列中出现的所有场景，即统计用户在预设期间内所经过的所有轨迹点。若用户第一天生活习惯轨迹向量序列为A、B、C、B、E、A；第二天生活习惯轨迹向量序列为A、B、D、B、F、A等，则该用户两天的生活习惯轨迹向量序列中出现的所有场景包括A、B、C、D、E和F等轨迹点(即场景)。可以理解地，生活习惯轨迹向量序列中出现的所有场景，可限定最终形成的马尔科夫转移矩阵的大小，即限定马尔科夫转移矩阵的行数和列数。
S32：计算任一场景与下一场景的转移概率。
对任一场景而言，从预设期间内所有生活习惯轨迹向量序列中获取与该场景相对应的下一场景，统计所有下一场景的总数和每个下一场景出现的次数，以计算任一场景与下一场景的转移概率，以便利用该转移概率构建马尔科夫转移矩阵。
S33：基于转移概率，构建马尔科夫转移矩阵。
在马尔科夫转移矩阵构建过程中，以生活习惯轨迹向量序列中出现的所有场景作为矩阵的行数和列数，即将所有场景分别作为t＝k时刻每一行的Xk场景，并将所有场景分别作为t＝k+1时刻每一列的Xk+1场景，其中，Xk+1场景是Xk场景的下一场景。在矩阵内分别填写每一Xk场景到Xk+1场景的转移概率，以构建马尔科夫转移矩阵。该马尔科夫转移矩阵可清楚显示用户在预测期间内所经过的所有场景，每一场景基于用户的地理位置信息获取，具有客观性和准确性，还可清楚显示从任一场景到下一场景的转移概率，所需采集的数据量小且预测结果准确性较高，可实现对用户行为进行较准确预测，以便于更好地开展业务推广等。
在一具体实施方式中，若用户1周内的生活习惯轨迹向量序列如下表所示：
周一 A B C D C E A
周二 A F L D C G A
周三 A B C H C E A
周四 A F C G L I A
周五 A B C H C G A
周六 A F A J E F A
周日 A B K G F G A
上表中，A为家庭住址，C为办公场所，B、D、E、F、G、H、I、K和L等对应除家庭住址和办公场所以外的其他活动场所，包括但不限于消费(包括吃饭消费)、娱乐、购物、健身等。上表中，用户在1周内的生活习惯轨迹向量序列中出现的所有场景包括A、B、C、D、E、F、G、H、I、J、K和L等12个，因此可构建12*12的马尔科夫转移矩阵。分别计算每一场景与下一场景的转移概率，以获取的马尔科夫转移矩阵如下所示。
Figure PCTCN2017076475-appb-000003
该马尔科夫转移矩阵可清楚显示从任一场景到下一场景的转移概率，所需采集的数据量小且预测结果准确性较高，可实现对用户行为场景进行较准确预测，以便于更好地开展业务推广等。
进一步地，在马尔科夫转移矩阵构建过程中，还可以使每一轨迹点与轨迹点所处时刻相关联，基于与所处时刻相关联的轨迹点构建马尔科夫转移矩阵，可进一步提高马尔科夫转移矩阵中对用户行为场景预测的准确性和可靠性。如统计用户在预设期间内的生活习惯轨迹向量序列中，以小时为单位，分别计算预设期间内同一时间范围内(如上午10点)内所有轨迹点及每一轨迹点的概率，基于时间顺序获取任一场景与下一场景的转换概率，并构建马尔科夫转移矩阵，使得形成的马尔科夫转移矩阵与轨迹点所处时刻相关联，进一步提高用户行为场景预测的准确性和可靠性。
S40：获取用户的当前场景，基于当前场景从马尔科夫转移矩阵中，获取对应的预测场景。
可以理解地，马尔科夫转移矩阵可清楚显示从任一场景到下一场景的转移概率，在任一时刻，获取用户的当前场景，即可从马尔科夫转移矩阵中获取其可能转移的所有下一场景以及每个下一场景的转移概率，根据转移概率的高低，选择转移概率较高的下一场景作为预测场景，以便基于获取到的预测场景对该用户开展业务推广活动。
本实施例所提供的个性化场景预测方法中，通过对用户在预设期间内获取的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列，由于地理位置信息具有较强的客观性和可 靠性，使其形成的生活习惯轨迹向量序列也具有较强的客观性和可靠性。基于生活习惯轨迹向量序列，构建马尔科夫转移矩阵，马尔科夫转移矩阵构建过程所需采集的数据量少，计算过程简单方便。由于马尔科夫转移矩阵可清楚显示从任一场景到下一场景的转移概率，使得基于马尔科夫转移矩阵获取预测场景时，所获取到的预测场景的准确性较高。
在一具体实施方式中，该个性化场景预测方法还包括如下步骤：
S50：基于马尔科夫转移矩阵，获取归一化转移矩阵。
其中，归一化转移矩阵是与多个马尔科夫转移矩阵具有高度相似性的矩阵，可将多个马尔科夫转移矩阵转换成归一化转移矩阵并存储，以实现节省存储空间的目的。
步骤S50具体包括如下步骤：
S51：获取多个用户的马尔科夫转移矩阵，每一马尔科夫转移矩阵与用户ID相关联。
其中，用户ID用于唯一识别用户，使用户ID与马尔科夫转移矩阵相关联，以实现通过用户ID确定马尔科夫转移矩阵对应的用户，以实现对该用户行为场景进行个性化预测。
S52：对多个马尔科夫转移矩阵进行逻辑回归处理，获取归一化转移矩阵。
即采用逻辑回归(Logistic Regression)模型对多个马尔科夫转移矩阵进行逻辑回归处理，以获取归一化转移矩阵，该归一化转移矩阵与多个马尔科夫转移矩阵具有高度相似性，可基于该归一化转移矩阵对用户行为场景预测，其预测效果与相对应的马尔科夫转移矩阵的预测效果相似，且归一化转换矩阵可大量节省存储空间。
S53：将归一化转移矩阵与多个用户ID关联存储。
可以理解地，将归一化转移矩阵与用户ID关联存储，即将构建归一化转移矩阵的多个马尔科夫转移矩阵对应的用户ID与该归一化转移矩阵关联存储，以实现基于任一用户ID可获取到其对应的归一化转移矩阵，并基于该归一化转移矩阵进行用户行为场景预测。将归一化转移矩阵与多个用户ID关联存储，无需存储多个用户ID对应的马尔科夫转移矩阵，可极大地节省存储空间。
在一具体实施方式中，该个性化场景预测方法还包括如下步骤：
S60：基于归一化转移矩阵进行场景预测。
由于归一化转移矩阵是与多个马尔科夫转移矩阵具有高度相似性的矩阵，基于归一化转移矩阵对用户行为预测时，其预测结果与采用马尔科夫转移矩阵对用户行为场景预测的预测结果也具有高度相似性，使得基于归一化转移矩阵对用户行为场景预测时，预测结果也具有较高的准确性和客观性。
步骤S60具体包括如下步骤：
S61：获取场景预测请求，场景预测请求包括用户ID和当前场景。
本实施例中，金融机构或者其他产品/服务提供者可向个性化场景预测设备输入所要进行行为预测的用户对应的用户ID，基于该用户ID对用户进行定位，以确定其对应的地理位置信息，从而确定用户的当前场景，以使个性化场景预测设备获取场景预测请求。
S62：基于场景预测请求中的用户ID，确定与用户ID相对应的相似用户的归一化转移矩阵。
可以理解地，归一化转移矩阵与多个用户ID关联存储，个性化场景预测设备基于获取到的场景预测请求可查询获取用户ID对应的归一化转移矩阵，以便于利用该归一化转移矩阵进行对用户进行场景预测。
S63：基于场景预测请求中的当前场景，从归一化转移矩阵中获取预测场景。
由于归一化转移矩阵是由多个用户的马尔科夫转移矩阵进行逻辑回归处理得到的，使得该归一化转移矩阵也可清楚显示任一场景到下一场景的转移概率，以便基于场景预测请求中的当前场景，获取其对应若干下一场景及每个下一场景的转移概率，以将转移概率较高的下一场景作为预测场景输出，以提高场景预测的准确性和客观性，并可达到节省存储空间的目的。
第二实施例
图3示出本实施例中个性化场景预测装置的原理框图。该个性化场景预测装置可以是金融机构或者其他产品/服务提供者上设置的个性化场景预测设备，用于实现对用户行为场景预测，以便于进行业务推广。如图3所示，该个性化场景预测装置包括处理器，处理器上设有位置信息获取模块10、轨迹向量序列获取模块20、转移矩阵构建模块30、预测场景获取模块40、归一化矩阵获取模块50和归一化场景预测模块60等功能模块。
位置信息获取模块10，用于基于位置服务获取用户的地理位置信息，地理位置信息包括与时间相关联的POI信息。
以任一用户一天的地理位置信息为例，该地理位置信息中包括0：00—24：00的POI信息，每一POI信息用于指示电子地图中的一点，包括POI点名称、经度和纬度等信息。基于用户的地理位置信息，可了解用户每天经过的家庭住址、办公场所、购物场所、娱乐场所、健身场所等信息。可以理解地，基于位置服务获取用户的地理位置信息，具有较强的客观性和可靠性。
基于位置服务(Location Based Service，简称LBS)是通过电信移动运营商的无线电通讯网络(如GSM网、CDMA网)或外部定位方式(如GPS)获取移动终端用户的位置信息(地 理坐标，或大地坐标)，在地理信息系统(Geographic Information System，简称GIS)平台的支持下，为用户提供相应服务的一种增值业务。总体来看，LBS由移动通信网络和计算机网络结合而成，两个网络之间通过网关实现交互。移动终端通过移动通信网络发出请求，经过网关传递给LBS服务平台；LBS服务平台根据用户请求和用户当前位置进行处理，并将结果通过网关返回给用户。POI(Point Of Interest,即兴趣点或信息点)，包括名称、类型、经度、纬度等资料，以使POI可在电子地图上呈现，以标示电子地图上的某个地点信息。
本实施例中，基于位置服务的移动终端为智能手机，通过开启智能手机上的定位功能，以使LBS服务平台实时获取智能手机的地理位置信息，从而了解携带该智能手机的用户的地理位置信息。该LBS服务平台与金融机构或者其他产品/服务提供者中的个性化场景预测设备相连，以使该个性化场景预测设备能够实时获取该智能手机对应的用户的地理位置信息。地理位置信息包括与时间相关联的POI信息中的时间包括日期和时刻，通过该地理位置信息可了解用户在任一时刻所处的POI信息。可以理解地，地理位置信息与用户ID相关联，用户ID用于识别唯一识别用户，可以是身份证号或手机号。
可以理解地，为了减少数据处理量，提高处理效率，可预先设置时间阈值，以使基于位置服务获取用户的地理位置信息时，只获取用户在任一地点停留时间达到该时间阈值的POI信息，以避免采集到的与时间相关联的POI信息的数据量较多，导致处理效率低的问题。
轨迹向量序列获取模块20，用于对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列。
其中，生活习惯轨迹向量序列由依据时间顺序排序的轨迹点组成。轨迹点是用户日常生活中经过的地点，可以是家庭住址、办公场所、购物场所、娱乐场所、健身场所等地点，可在电子地图中显示。其中，预设期间可以是当前系统时间之前的任意一段时间，可以为一周、一个月、三个月或半年，可根据需求自主设置。可以理解地，预设期间越长，其采集到的地理位置信息的数据量越多，处理结果的准确性越高；预设期间越短，其处理效率越高。为说明本实施例所提供的个性化场景预设装置的实现用户行为场景预测过程，可将预设期间设为1周，以便于计算。
进一步地，轨迹向量序列获取模块20具体包括子集群获取单元21、轨迹点获取单元22和向量序列获取单元23。
子集群获取单元21，用于采用DBSCAN算法对任一用户在预设期间内所有POI信息进行聚类，以获取若干子集群。
其中，DBSCAN(Density-Based Spatial Clustering of Applications with Noise，具有噪声的基于密度的聚类方法)是一种基于密度的空间算法。该算法将具有足够密度的区域划分为簇，并在具有噪声的空间数据库中发现任意形状的簇，它将簇定义为密度相连的点的最大集合。DBSCAN算法具有聚类速度快且能够有效处理噪声和发现任意形成的空间聚类的优点。
本实施例中，预先设置DBSCAN算法中的预设扫描半径(以下简称为eps)和最小包含点数(minPts)，任选一个未被访问(unvisited)的POI信息开始，找出与其距离在eps之内(包括eps)的所有POI信息，将POI信息与距离在eps之内的所有POI信息作为一个子集群输出。
轨迹点获取单元22，用于采用K-MEANS算法对每一子集群进行迭代聚合，获取每一子集群的质心POI信息，并将质心POI信息作为轨迹点输出。
K-MEANS算法是很典型的基于距离的算法，采用距离作为相似性的评价指标，即认为两个对象的距离越近，其相似度就越大。其计算公式为
Figure PCTCN2017076475-appb-000004
其中，k个初始类聚类中心点的选取对聚类结果具有较大的影响，因为在该算法第一步中是随机的选取任意k个对象作为初始聚类的中心，初始地代表一个簇。该算法在每次迭代中对数据集中剩余的每个对象，根据其与各个簇中心的距离将每个对象重新赋给最近的簇。若一次迭代前后，J的值没有发生变化，说明算法已经收敛。K-MEANS算法可快速简单地对数据进行聚类，对大数据集具有较高的效率且可伸缩性，时间复杂度近于线性，而且适合挖掘大规模数据集。
本实施例中，采用K-MEANS算法对每一子集群中的POI信息进行迭代聚合，直到最后一次迭代时，迭代前后数值没有发生变化，则获取该子集群的质心POI信息，该质心POI信息对应一轨迹点。
向量序列获取单元23，用于基于轨迹点的时间序列，确定用户在预设期间内的生活习惯轨迹向量序列。
本实施例中，通过对用户在预设期间内每日采集到的地理位置信息进行聚类分析，获取每日由按时间顺序排序的轨迹点形成的生活习惯轨迹向量序列。该生活习惯轨迹向量序列可清楚体现用户每日经过的家庭住址、办公场所、购物场所、娱乐场所、健身场所等轨迹点，具有较强的客观性和可靠性。
在一具体实施方式中，若A为家庭住址，B为办公场所，C为购物场所，D为娱乐场所，E为健身场所，F为公园，G为医院等等；且A’和A”为A附近500m内的地点，B’和B” 为B附近500m内的地点，C’和C”为C附近500m内的地点，D’和D”为D附近500m内的地点，D’和D”为D附近500m内的地点，D’和D”为D附近500m内的地点，……G’和G”为G附近500m内的地点。在1周内，第一天的地理位置信息包括A、A’、B’、B、C”、C、B”、B、E”、E、A”、A等POI信息；第二天的地理位置信息包括A、A’、B’、B、D”、D、B”、B、F”、F、A”、A等POI信息……依此类推。步骤S21中采用DBSCAN算法进行聚类时，将1周内所有POI信息，通过设置扫描半径(eps)为500m和最小包含点数(minPts)为1，以将A、A’、A”作为一子集群输出，将B，B’、B”作为一子集群输出……G，G’、G”作为一子集群输出。步骤S22中采用K-MEANS算法对每一子集群进行聚类，获取到子集群中的质心POI信息，对于子集群A、A’、A”而言，采用K-MEANS算法进行迭代聚类时，获取到的质心POI信息为A，将A作为轨迹点输出，依此类推，获取其他轨迹点B、C、D、E、F和G。本实施例中，任一子集群中质心POI信息出现的频率大于其他POI信息出现的频率。步骤S23中，基于轨迹点的时间序列，获取用户在预设期间内每日的生活习惯轨迹向量序列，如第一天的轨迹点为A、B、C、B、E、A，第二天的轨迹点为A、B、D、B、F、A……等。
转移矩阵构建模块30，用于基于生活习惯轨迹向量序列，构建马尔科夫转移矩阵。
马尔科夫转移矩阵，即马尔科夫(Markov Process)链的转移概率(transition probability)矩阵，是一种利用马尔科夫分析方法基于概率建立的随机型的时序模型。马尔科夫分析方法的基本模型为：X(k+1)＝X(k)*P，其中，X(k)为预测用户在t＝k时刻的场景向量，P表示一步转移概率矩阵；X(k+1)为预测用户在t＝k+1时刻的场景向量。本实施例中，所获取的马尔科夫转移矩阵如下：
Figure PCTCN2017076475-appb-000005
进一步地，转移矩阵构建模块30具体包括场景获取单元31、概率计算单元32和矩阵构建单元33。
场景获取单元31，用于基于生活习惯轨迹向量序列，获取生活习惯轨迹向量序列中出现的所有场景。
本实施例中，生活习惯轨迹向量序列中的每一轨迹点对应一场景。基于生活习惯轨迹向量序列，获取生活习惯轨迹向量序列中出现的所有场景，即统计用户在预设期间内所经过的所有轨迹点。若用户第一天生活习惯轨迹向量序列为A、B、C、B、E、A；第二天生活习惯轨迹向量序列为A、B、D、B、F、A等，则该用户两天的生活习惯轨迹向量序列中出现的所有场景包括A、B、C、D、E和F等轨迹点(即场景)。可以理解地，生活习惯轨迹向量序列中出现的所有场景，可限定最终形成的马尔科夫转移矩阵的大小，即限定马尔科夫转移矩阵的行数和列数。
概率计算单元32，用于计算任一场景与下一场景的转移概率。
对任一场景而言，从预设期间内所有生活习惯轨迹向量序列中获取与该场景相对应的下一场景，统计所有下一场景的总数和每个下一场景出现的次数，以计算任一场景与下一场景的转移概率，以便利用该转移概率构建马尔科夫转移矩阵。
矩阵构建单元33，用于基于转移概率，构建马尔科夫转移矩阵。
在马尔科夫转移矩阵构建过程中，以生活习惯轨迹向量序列中出现的所有场景作为矩阵的行数和列数，即将所有场景分别作为t＝k时刻每一行的Xk场景，并将所有场景分别作为t＝k+1时刻每一列的Xk+1场景，其中，Xk+1场景是Xk场景的下一场景。在矩阵内分别填写每一Xk场景到Xk+1场景的转移概率，以构建马尔科夫转移矩阵。该马尔科夫转移矩阵可清楚显示用户在预测期间内所经过的所有场景，每一场景基于用户的地理位置信息获取，具有客观性和准确性，还可清楚显示从任一场景到下一场景的转移概率，所需采集的数据量小且预测结果准确性较高，可实现对用户行为进行较准确预测，以便于更好地开展业务推广等。
在一具体实施方式中，若用户1周内的生活习惯轨迹向量序列如下表所示：
周一 A B C D C E A
周二 A F L D C G A
周三 A B C H C E A
周四 A F C G L I A
周五 A B C H C G A
周六 A F A J E F A
周日 A B K G F G A
上表中，A为家庭住址，C为办公场所，B、D、E、F、G、H、I、K和L等对应除家庭住址和办公场所以外的其他活动场所，包括但不限于消费(包括吃饭消费)、娱乐、购物、 健身等。上表中，用户在1周内的生活习惯轨迹向量序列中出现的所有场景包括A、B、C、D、E、F、G、H、I、J、K和L等12个，因此可构建12*12的马尔科夫转移矩阵。分别计算每一场景与下一场景的转移概率，以获取的马尔科夫转移矩阵如下所示。
Figure PCTCN2017076475-appb-000006
该马尔科夫转移矩阵可清楚显示从任一场景到下一场景的转移概率，所需采集的数据量小且预测结果准确性较高，可实现对用户行为场景进行较准确预测，以便于更好地开展业务推广等。
进一步地，在马尔科夫转移矩阵构建过程中，还可以使每一轨迹点与轨迹点所处时刻相关联，基于与所处时刻相关联的轨迹点构建马尔科夫转移矩阵，可进一步提高马尔科夫转移矩阵中对用户行为场景预测的准确性和可靠性。如统计用户在预设期间内的生活习惯轨迹向量序列中，以小时为单位，分别计算预设期间内同一时间范围内(如上午10点)内所有轨迹点及每一轨迹点的概率，基于时间顺序获取任一场景与下一场景的转换概率，并构建马尔科夫转移矩阵，使得形成的马尔科夫转移矩阵与轨迹点所处时刻相关联，进一步提高用户行为场景预测的准确性和可靠性。
预测场景获取模块40，用于获取用户的当前场景，基于当前场景从马尔科夫转移矩阵中，获取对应的预测场景。
可以理解地，马尔科夫转移矩阵可清楚显示从任一场景到下一场景的转移概率，在任一时刻，获取用户的当前场景，即可从马尔科夫转移矩阵中获取其可能转移的所有下一场景以及每个下一场景的转移概率，根据转移概率的高低，选择转移概率较高的下一场景作为预 测场景，以便基于获取到的预测场景对该用户开展业务推广活动。
本实施例所提供的个性化场景预测装置中，通过对用户在预设期间内获取的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列，由于地理位置信息具有较强的客观性和可靠性，使其形成的生活习惯轨迹向量序列也具有较强的客观性和可靠性。基于生活习惯轨迹向量序列，构建马尔科夫转移矩阵，马尔科夫转移矩阵构建过程所需采集的数据量少，计算过程简单方便。由于马尔科夫转移矩阵可清楚显示从任一场景到下一场景的转移概率，使得基于马尔科夫转移矩阵获取预测场景时，所获取到的预测场景的准确性较高。
在一具体实施方式中，该个性化场景预测装置还包括归一化矩阵获取模块50，用于基于马尔科夫转移矩阵，获取归一化转移矩阵。
其中，归一化转移矩阵是与多个马尔科夫转移矩阵具有高度相似性的矩阵，可将多个马尔科夫转移矩阵转换成归一化转移矩阵并存储，以实现节省存储空间的目的。
归一化矩阵获取模块50具体包括矩阵获取单元51、逻辑回归处理单元52和矩阵关联存储单元53。
矩阵获取单元51，用于获取多个用户的马尔科夫转移矩阵，每一马尔科夫转移矩阵与用户ID相关联。
其中，用户ID用于唯一识别用户，使用户ID与马尔科夫转移矩阵相关联，以实现通过用户ID确定马尔科夫转移矩阵对应的用户，以实现对该用户行为场景进行个性化预测。
逻辑回归处理单元52，用于对多个马尔科夫转移矩阵进行逻辑回归处理，获取归一化转移矩阵。
即采用逻辑回归(Logistic Regression)模型对多个马尔科夫转移矩阵进行逻辑回归处理，以获取归一化转移矩阵，该归一化转移矩阵与多个马尔科夫转移矩阵具有高度相似性，可基于该归一化转移矩阵对用户行为场景预测，其预测效果与相对应的马尔科夫转移矩阵的预测效果相似，且归一化转换矩阵可大量节省存储空间。
矩阵关联存储单元53，用于将归一化转移矩阵与多个用户ID关联存储。
可以理解地，将归一化转移矩阵与用户ID关联存储，即将构建归一化转移矩阵的多个马尔科夫转移矩阵对应的用户ID与该归一化转移矩阵关联存储，以实现基于任一用户ID可获取到其对应的归一化转移矩阵，并基于该归一化转移矩阵进行用户行为场景预测。将归一化转移矩阵与多个用户ID关联存储，无需存储多个用户ID对应的马尔科夫转移矩阵，可极大地节省存储空间。
在一具体实施方式中，该个性化场景预测装置还包括归一化场景预测模块60，用于基 于归一化转移矩阵进行场景预测。
由于归一化转移矩阵是与多个马尔科夫转移矩阵具有高度相似性的矩阵，基于归一化转移矩阵对用户行为预测时，其预测结果与采用马尔科夫转移矩阵对用户行为场景预测的预测结果也具有高度相似性，使得基于归一化转移矩阵对用户行为场景预测时，预测结果也具有较高的准确性和客观性。
归一化场景预测模块60具体包括预测请求获取单元61、归一化矩阵获取单元62和归一化场景预测单元63。
预测请求获取单元61，用于获取场景预测请求，场景预测请求包括用户ID和当前场景。
本实施例中，金融机构或者其他产品/服务提供者可向个性化场景预测设备输入所要进行行为预测的用户对应的用户ID，基于该用户ID对用户进行定位，以确定其对应的地理位置信息，从而确定用户的当前场景，以使个性化场景预测设备获取场景预测请求。
归一化矩阵获取单元62，用于基于场景预测请求中的用户ID，确定与用户ID相对应的相似用户的归一化转移矩阵。
可以理解地，归一化转移矩阵与多个用户ID关联存储，个性化场景预测设备基于获取到的场景预测请求可查询获取用户ID对应的归一化转移矩阵，以便于利用该归一化转移矩阵进行对用户进行场景预测。
归一化场景预测单元63，用于基于场景预测请求中的当前场景，从归一化转移矩阵中获取预测场景。
由于归一化转移矩阵是由多个用户的马尔科夫转移矩阵进行逻辑回归处理得到的，使得该归一化转移矩阵也可清楚显示任一场景到下一场景的转移概率，以便基于场景预测请求中的当前场景，获取其对应若干下一场景及每个下一场景的转移概率，以将转移概率较高的下一场景作为预测场景输出，以提高场景预测的准确性和客观性，并可达到节省存储空间的目的。
第三实施例
图3是本发明第三实施例的个性化场景预测设备300的结构示意图。其中，设备300可以为手机、平板电脑、个人数字助理(PersonalDigital Assistant，PDA)和或车载电脑等具有一定数据处理能力的移动终端、或者台式电脑、服务器等终端。如图3所示，设备300包括射频(RadioFrequency，RF)电路301、存储器302、输入模块303、显示模块304、处理器305、音频电路306、WiFi(WirelessFidelity)模块307和电源308。
输入模块303和显示模块304作为设备300的用户交互装置，用于实现用户与设备300之间的交互，例如，接收用户输入的场景预测指令并显示对应的预测场景。输入模块303用于接收用户输入的场景预测指令，并将所述场景预测指令发送给所述处理器305，所述场景预测指令包括所述当前场景。所述处理器305用于接收所述场景预测指令，并基于所述场景预测指令，获取所述预测场景，并将所述预测场景发送给所述显示模块304。显示模块304接收并显示预测场景。
在一些实施例中，输入模块303可用于接收用户输入的数字或字符信息，以及产生与设备300的用户设置以及功能控制有关的信号输入。在一些实施例中，该输入模块303可以包括触控面板3031。触控面板3031，也称为触摸屏，可收集用户在其上或附近的触摸操作(比如用户使用手指、触笔等任何适合的物体或附件在触控面板3031上的操作)，并根据预先设定的程式驱动相应的连接装置。可选地，触控面板3031可包括触摸检测装置和触摸控制器两个部分。其中，触摸检测装置检测用户的触摸方位，并检测触摸操作带来的信号，将信号传送给触摸控制器；触摸控制器从触摸检测装置上接收触摸信息，并将它转换成触点坐标，再送给该处理器305，并能接收处理器305发来的命令并加以执行。此外，可以采用电阻式、电容式、红外线以及表面声波等多种类型实现触控面板3031。除了触控面板3031，输入模块303还可以包括其他输入设备3032，其他输入设备3032可以包括但不限于物理键盘、功能键(比如音量控制按键、开关按键等)、轨迹球、鼠标、操作杆等中的一种或多种。
在一些实施例中，显示模块304可用于显示由用户输入的信息或提供给用户的信息以及设备300的各种菜单界面。显示模块304可包括显示面板3041，可选地，可以采用LCD或有机发光二极管(Organic Light-Emitting Diode，OLED)等形式来配置显示面板3041。
可以理解地，触控面板3031可以覆盖显示面板3041，形成触摸显示屏，当该触摸显示屏检测到在其上或附近的触摸操作后，传送给处理器305以确定触摸事件的类型，随后处理器305根据触摸事件的类型在触摸显示屏上提供相应的视觉输出。
触摸显示屏包括应用程序界面显示区及常用控件显示区。该应用程序界面显示区及该常用控件显示区的排列方式并不限定，可以为上下排列、左右排列等可以区分两个显示区的排列方式。该应用程序界面显示区可以用于显示应用程序的界面。每一个界面可以包含至少一个应用程序的图标和/或widget桌面控件等界面元素。该应用程序界面显示区也可以为不包含任何内容的空界面。该常用控件显示区用于显示使用率较高的控件，例如，设置按钮、界面编号、滚动条、电话本图标等应用程序图标等。
WiFi模块307作为设备300的网络接口，可以实现设备300与其他设备的数据交互， 本实施例中，网络接口可与远端存储设备和外部显示设备通过网络通信相连。所述网络接口用于接收所述远端存储设备发送的所述地理位置信息和所述当前场景，并将所述地理位置信息和所述当前场景发送给所述处理器305；还用于接收所述处理器305发送的所述预测场景，并将所述预测场景发送给所述外部显示设备。外部显示设备可接收并显示所述预测场景。本实施例中，与该网络接口通过WiFi网络相连的远端存储设备可以是云服务器或其他数据库，该远端存储设备上存储有所述地理位置信息和所述当前场景，可将所述地理位置信息和所述当前场景可通过WiFi网络发送给WiFi模块307，WiFi模块307将获取到的所述地理位置围栏信息和所述当前场景发送给所述处理器305，并将所述预测场景发送给所述外部显示设备。
存储器302包括第一存储器3021及第二存储器3022。在一些实施例中，第一存储器3021可为非易失性计算机可读存储介质，其上存储有操作系统、数据库及计算机可执行指令。计算机可执行指令可被处理器305所执行，用于实现如图1所示的实施例的个性化场景预测方法。存储器302上的数据库用于存储各类数据，例如，上述个性化场景预测方法中所涉及的各种数据，如地理位置信息和所述马尔科夫转移矩阵。第二存储器3021可为设备300的内存储器，为非易失性计算机可读存储介质中的操作系统、数据库和计算机可执行指令提供高速缓存的运行环境。
在本实施例中，处理器305是设备300的控制中心，利用各种接口和线路连接整个手机的各个部分，通过运行或执行存储在第一存储器3021内的计算机可执行搜集和/或数据库内的数据，执行设备300的各种功能和处理数据，从而对设备300进行整体监控。可选地，处理器305可包括一个或多个处理模块。
在本实施例中，通过执行存储该第一存储器3021内的计算机可执行指令和/或数据库内的数据，所述处理器305用于执行如下步骤：基于位置服务获取用户的地理位置信息，所述地理位置信息包括与时间相关联的POI信息；对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列；基于所述生活习惯轨迹向量序列，构建马尔科夫转移矩阵；获取用户的当前场景，基于所述当前场景从所述马尔科夫转移矩阵中，获取对应的预测场景。
优选地，所述对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列，包括：
采用DBSCAN算法对任一用户在预设期间内所有POI信息进行聚类，以获取若干子集群；
采用K-MEANS算法对每一所述子集群进行迭代聚合，获取每一所述子集群的质心POI 信息，并将所述质心POI信息作为轨迹点输出；
基于所述轨迹点的时间序列，确定所述用户在所述预设期间内的生活习惯轨迹向量序列。
优选地，所述基于所述生活习惯轨迹向量序列，构建马尔科夫转移矩阵，包括：
基于所述生活习惯轨迹向量序列，获取所述生活习惯轨迹向量序列中出现的所有场景；
计算任一场景与下一场景的转移概率；
基于所述转移概率，构建所述马尔科夫转移矩阵。
优选地，所述处理器305还执行如下步骤：基于所述马尔科夫转移矩阵，获取归一化转移矩阵；
所述基于所述马尔科夫转移矩阵，获取归一化转移矩阵包括：
获取多个用户的马尔科夫转移矩阵，每一马尔科夫转移矩阵与用户ID相关联；
对多个所述马尔科夫转移矩阵进行逻辑回归处理，获取归一化转移矩阵；
将所述归一化转移矩阵与多个所述用户ID关联存储。
优选地，所述处理器305还执行如下步骤：基于所述归一化转移矩阵进行场景预测；
所述基于所述归一化转移矩阵进行场景预测，包括：
获取场景预测请求，所述场景预测请求包括用户ID和当前场景；
基于所述场景预测请求中的用户ID，确定与用户ID相对应的归一化转移矩阵；
基于所述场景预测请求中的当前场景，从所述归一化转移矩阵中获取预测场景。
本发明实施例的设备300，处理器305可基于位置服务获取用户的地理位置信息，所述地理位置信息包括与时间相关联的POI信息；对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列；基于所述生活习惯轨迹向量序列，构建马尔科夫转移矩阵；获取用户的当前场景，基于所述当前场景从所述马尔科夫转移矩阵中，获取对应的预测场景，使得基于设备300可对用户场景进行预测，且所获取的预测场景的准确性较高，客观性和可靠性强，计算过程简单方便。进一步地，处理器305对地理位置信息进行聚类分析获取的生活经验习惯轨迹向量序列可以存储在处理器缓存中，也可以存储在存储器302中；处理器305可将基于当前场景获取的预测场景发送给用户交互装置(如显示单元304)或者网络接口(如WiFi模块307)。可以理解地，显示单元304可显示该预测场景；WiFi模块307可将该预测场景发送给外部显示设备，由外部显示设备显示该预测场景。
第四实施例
本实施例提供一种非易失性计算机可读存储介质。该非易失性计算机可读存储介质用 于存储一个或多个计算机可执行指令。具体地，计算机可执行指令被一个或多个处理器执行，使得所述一个或多个处理器执行第一实施例所述个性化场景预测方法，为避免重复，这里不再赘述。
本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的模块及算法步骤，能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本发明的范围。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统、装置和模块的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在本申请所提供的实施例中，应该理解到，所揭露的装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述模块的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个模块或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或模块的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的模块可以是或者也可以不是物理上分开的，作为模块显示的部件可以是或者也可以不是物理模块，即可以位于一个地方，或者也可以分布到多个网络模块上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。
另外，在本发明各个实施例中的各功能模块可以集成在一个处理模块中，也可以是各个模块单独物理存在，也可以两个或两个以上模块集成在一个模块中。
所述功能如果以软件功能模块的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本发明的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本发明各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、ROM、RAM、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述，仅为本发明的具体实施方式，但本发明的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本发明揭露的技术范围内，可轻易想到变化或替换，都应涵 盖在本发明的保护范围之内。因此，本发明的保护范围应以权利要求的保护范围为准。

Claims (19)

  1. 一种个性化场景预测方法，其特征在于，包括：
    基于位置服务获取用户的地理位置信息，所述地理位置信息包括与时间相关联的POI信息；
    对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列；
    基于所述生活习惯轨迹向量序列，构建马尔科夫转移矩阵；
    获取用户的当前场景，基于所述当前场景从所述马尔科夫转移矩阵中，获取对应的预测场景。
  2. 根据权利要求1所述的个性化场景预测方法，其特征在于，所述对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列，包括：
    采用DBSCAN算法对任一用户在预设期间内所有POI信息进行聚类，以获取若干子集群；
    采用K-MEANS算法对每一所述子集群进行迭代聚合，获取每一所述子集群的质心POI信息，并将所述质心POI信息作为轨迹点输出；
    基于所述轨迹点的时间序列，确定所述用户在所述预设期间内的生活习惯轨迹向量序列。
  3. 根据权利要求2所述的个性化场景预测方法，其特征在于，所述基于所述生活习惯轨迹向量序列，构建马尔科夫转移矩阵，包括：
    基于所述生活习惯轨迹向量序列，获取所述生活习惯轨迹向量序列中出现的所有场景；
    计算任一场景与下一场景的转移概率；
    基于所述转移概率，构建所述马尔科夫转移矩阵。
  4. 根据权利要求1所述的个性化场景预测方法，其特征在于，还包括：基于所述马尔科夫转移矩阵，获取归一化转移矩阵；
    所述基于所述马尔科夫转移矩阵，获取归一化转移矩阵，包括：
    获取多个用户的马尔科夫转移矩阵，每一马尔科夫转移矩阵与用户ID相关联；
    对多个所述马尔科夫转移矩阵进行逻辑回归处理，获取归一化转移矩阵；
    将所述归一化转移矩阵与多个所述用户ID关联存储。
  5. 根据权利要求4所述的个性化场景预测方法，其特征在于，还包括：基于所述归一化转移矩阵进行场景预测；
    所述基于所述归一化转移矩阵进行场景预测，包括：
    获取场景预测请求，所述场景预测请求包括用户ID和当前场景；
    基于所述场景预测请求中的用户ID，确定与用户ID相对应的归一化转移矩阵；
    基于所述场景预测请求中的当前场景，从所述归一化转移矩阵中获取预测场景。
  6. 一种个性化场景预测装置，其特征在于，包括：
    位置信息获取模块，用于基于位置服务获取用户的地理位置信息，所述地理位置信息包括与时间相关联的POI信息；
    轨迹向量序列获取模块，用于对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列；
    转移矩阵构建模块，用于基于所述生活习惯轨迹向量序列，构建马尔科夫转移矩阵；
    预测场景获取模块，用于获取用户的当前场景，基于所述当前场景从所述马尔科夫转移矩阵中，获取对应的预测场景。
  7. 根据权利要求6所述的个性化场景预测装置，其特征在于，所述轨迹向量序列获取模块包括：
    子集群获取单元，用于采用DBSCAN算法对任一用户在预设期间内所有POI信息进行聚类，以获取若干子集群；
    轨迹点获取单元，用于采用K-MEANS算法对每一所述子集群进行迭代聚合，获取每一所述子集群的质心POI信息，并将所述质心POI信息作为轨迹点输出；
    向量序列获取单元，用于基于所述轨迹点的时间序列，确定所述用户在所述预设期间内的生活习惯轨迹向量序列。
  8. 根据权利要求7所述的个性化场景预测装置，其特征在于，所述转移矩阵构建模块包括：
    场景获取单元，用于基于所述生活习惯轨迹向量序列，获取所述生活习惯轨迹向量序列中出现的所有场景；
    概率计算单元，用于计算任一场景与下一场景的转移概率；
    矩阵构建单元，用于基于所述转移概率，构建所述马尔科夫转移矩阵。
  9. 根据权利要求6所述的个性化场景预测装置，其特征在于，还包括归一化矩阵获取模块，用于基于所述马尔科夫转移矩阵，获取归一化转移矩阵；
    所述归一化矩阵获取模块包括：
    矩阵获取单元，用于获取多个用户的马尔科夫转移矩阵，每一马尔科夫转移矩阵与用 户ID相关联；
    逻辑回归处理单元，用于对多个所述马尔科夫转移矩阵进行逻辑回归处理，获取归一化转移矩阵；
    矩阵关联存储单元，用于将所述归一化转移矩阵与多个所述用户ID关联存储。
  10. 根据权利要求9所述的个性化场景预测装置，其特征在于，还包括：归一化场景预测模块，用于基于所述归一化转移矩阵进行场景预测；
    所述归一化场景预测模块，包括：
    预测请求获取单元，用于获取场景预测请求，所述场景预测请求包括用户ID和当前场景；
    归一化矩阵获取单元，用于基于所述场景预测请求中的用户ID，确定与用户ID相对应的归一化转移矩阵；
    归一化场景预测单元，用于基于所述场景预测请求中的当前场景，从所述归一化转移矩阵中获取预测场景。
  11. 一种个性化场景预测设备，其特征在于，包括处理器及存储器，所述存储器存储有计算机可执行指令，所述处理器用于执行所述计算机可执行指令以执行如下步骤：
    基于位置服务获取用户的地理位置信息，所述地理位置信息包括与时间相关联的POI信息；
    对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列；
    基于所述生活习惯轨迹向量序列，构建马尔科夫转移矩阵；
    获取用户的当前场景，基于所述当前场景从所述马尔科夫转移矩阵中，获取对应的预测场景。
  12. 根据权利要求11所述的设备，其特征在于，所述对用户在预设期间内所有的地理位置信息进行聚类分析，获取生活习惯轨迹向量序列，包括：
    采用DBSCAN算法对任一用户在预设期间内所有POI信息进行聚类，以获取若干子集群；
    采用K-MEANS算法对每一所述子集群进行迭代聚合，获取每一所述子集群的质心POI信息，并将所述质心POI信息作为轨迹点输出；
    基于所述轨迹点的时间序列，确定所述用户在所述预设期间内的生活习惯轨迹向量序列。
  13. 根据权利要求12所述的设备，其特征在于，所述基于所述生活习惯轨迹向量序列， 构建马尔科夫转移矩阵，包括：
    基于所述生活习惯轨迹向量序列，获取所述生活习惯轨迹向量序列中出现的所有场景；
    计算任一场景与下一场景的转移概率；
    基于所述转移概率，构建所述马尔科夫转移矩阵。
  14. 根据权利要求11所述的设备，其特征在于，所述处理器还执行如下步骤：基于所述马尔科夫转移矩阵，获取归一化转移矩阵；
    所述基于所述马尔科夫转移矩阵，获取归一化转移矩阵，包括：
    获取多个用户的马尔科夫转移矩阵，每一马尔科夫转移矩阵与用户ID相关联；
    对多个所述马尔科夫转移矩阵进行逻辑回归处理，获取归一化转移矩阵；
    将所述归一化转移矩阵与多个所述用户ID关联存储。
  15. 根据权利要求14所述的设备，其特征在于，所述处理器还执行如下步骤：基于所述归一化转移矩阵进行场景预测；
    所述基于所述归一化转移矩阵进行场景预测，包括：
    获取场景预测请求，所述场景预测请求包括用户ID和当前场景；
    基于所述场景预测请求中的用户ID，确定与用户ID相对应的归一化转移矩阵；
    基于所述场景预测请求中的当前场景，从所述归一化转移矩阵中获取预测场景。
  16. 根据权利要求11所述的设备，其特征在于，所述设备还包括与所述处理器相连的网络接口；所述网络接口与远端存储设备和外部显示设备相连；所述网络接口用于接收所述远端存储设备发送的所述地理位置信息和所述当前场景，并将所述地理位置信息和所述当前场景发送给所述处理器；还用于接收所述处理器发送的所述预测场景，并将所述预测场景发送给所述外部显示设备。
  17. 根据权利要求11所述的设备，其特征在于，所述设备还包括与所述处理器相连的用户交互装置，所述用户交互装置用于接收用户输入的场景预测指令并将所述场景预测指令发送给所述处理器；所述场景预测指令包括所述当前场景；
    所述处理器，用于基于所述场景预测指令，获取所述预测场景，并将所述预测场景发送给所述用户交互装置；
    所述用户交互装置，还用于接收并显示所述预测场景。
  18. 根据权利要求11所述的设备，其特征在于，所述存储器中存储有数据库，用于存储所述地理位置信息和所述马尔科夫转移矩阵。
  19. 一种非易失性计算机可读存储介质，其特征在于，用于存储一个或多个计算机可 执行指令，所述计算机可执行指令被一个或多个处理器执行，使得所述一个或多个处理器执行权利要求1-5任一项所述个性化场景预测方法。
PCT/CN2017/076475 2016-12-29 2017-03-13 个性化场景预测方法、装置、设备和存储介质 WO2018120428A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201611249507.6 2016-12-29
CN201611249507.6A CN106709606B (zh) 2016-12-29 2016-12-29 个性化场景预测方法及装置

Publications (1)

Publication Number Publication Date
WO2018120428A1 true WO2018120428A1 (zh) 2018-07-05

Family

ID=58903991

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/076475 WO2018120428A1 (zh) 2016-12-29 2017-03-13 个性化场景预测方法、装置、设备和存储介质

Country Status (2)

Country Link
CN (1) CN106709606B (zh)
WO (1) WO2018120428A1 (zh)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111064617A (zh) * 2019-12-16 2020-04-24 重庆邮电大学 基于经验模态分解聚类的网络流量预测方法及装置
CN111161042A (zh) * 2019-11-26 2020-05-15 深圳壹账通智能科技有限公司 个人风险评估方法、装置、终端及存储介质
CN111738323A (zh) * 2020-06-12 2020-10-02 上海应用技术大学 基于灰色马尔科夫模型的混合增强智能轨迹预测方法和装置
CN111753386A (zh) * 2019-03-11 2020-10-09 北京嘀嘀无限科技发展有限公司 一种数据处理方法及装置
CN111798260A (zh) * 2019-04-09 2020-10-20 Oppo广东移动通信有限公司 用户行为预测模型构建方法、装置、存储介质及电子设备
CN111796314A (zh) * 2019-04-09 2020-10-20 Oppo广东移动通信有限公司 信息处理方法、装置、存储介质及电子设备
CN112560910A (zh) * 2020-12-02 2021-03-26 中国联合网络通信集团有限公司 用户分类方法和装置
CN113938817A (zh) * 2021-09-10 2022-01-14 哈尔滨工业大学（威海） 一种基于车辆位置信息的车主出行位置预测方法
CN115515143A (zh) * 2021-06-22 2022-12-23 中国移动通信集团重庆有限公司 一种基站覆盖场景确定方法、装置、设备及存储介质
CN117035176A (zh) * 2023-08-09 2023-11-10 上海智租物联科技有限公司 基于马尔可夫链的用户换电行为预测与定向召回方法
CN117370825A (zh) * 2023-10-11 2024-01-09 国网经济技术研究院有限公司 基于注意力条件生成对抗网络的长期场景生成方法及系统
CN117609737A (zh) * 2024-01-18 2024-02-27 中国人民解放军火箭军工程大学 一种惯性导航系统健康状态预测方法、系统、设备及介质

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109413683B (zh) 2017-08-15 2021-09-21 华为技术有限公司 一种获取发射概率、转移概率以及序列定位的方法和装置
CN107590244B (zh) * 2017-09-14 2020-04-17 深圳市和讯华谷信息技术有限公司 移动设备线下活动场景的识别方法和装置
CN108304510B (zh) * 2018-01-19 2021-04-16 武汉大学 公共地图服务用户城市访问行为与poi类型关联规则的提取方法
CN109460509A (zh) * 2018-10-12 2019-03-12 平安科技（深圳）有限公司 用户兴趣点评估方法、装置、计算机设备及存储介质
CN111159318A (zh) * 2018-11-08 2020-05-15 阿里巴巴集团控股有限公司 聚合兴趣点的方法、装置、设备和介质
CN114201286B (zh) * 2022-02-16 2022-04-26 成都明途科技有限公司 一种任务的处理方法、装置、电子设备、存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103731916A (zh) * 2014-01-14 2014-04-16 上海河广信息科技有限公司 一种基于无线网络的用户位置预测系统和方法
CN103747523A (zh) * 2014-01-14 2014-04-23 上海河广信息科技有限公司 一种基于无线网络的用户位置预测系统和方法
JP5641243B2 (ja) * 2011-09-06 2014-12-17 トヨタ自動車東日本株式会社 作業支援システム
CN104239556A (zh) * 2014-09-25 2014-12-24 西安理工大学 基于密度聚类的自适应轨迹预测方法
CN106028444A (zh) * 2016-07-01 2016-10-12 国家计算机网络与信息安全管理中心 预测移动终端位置的方法及装置

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9846049B2 (en) * 2008-07-09 2017-12-19 Microsoft Technology Licensing, Llc Route prediction
CN105183800A (zh) * 2015-08-25 2015-12-23 百度在线网络技术（北京）有限公司 信息预测的方法和装置
CN105488213B (zh) * 2015-12-11 2019-05-24 哈尔滨工业大学深圳研究生院 面向lbs的基于马尔可夫预测算法的个性化推荐方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP5641243B2 (ja) * 2011-09-06 2014-12-17 トヨタ自動車東日本株式会社 作業支援システム
CN103731916A (zh) * 2014-01-14 2014-04-16 上海河广信息科技有限公司 一种基于无线网络的用户位置预测系统和方法
CN103747523A (zh) * 2014-01-14 2014-04-23 上海河广信息科技有限公司 一种基于无线网络的用户位置预测系统和方法
CN104239556A (zh) * 2014-09-25 2014-12-24 西安理工大学 基于密度聚类的自适应轨迹预测方法
CN106028444A (zh) * 2016-07-01 2016-10-12 国家计算机网络与信息安全管理中心 预测移动终端位置的方法及装置

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111753386B (zh) * 2019-03-11 2024-03-26 北京嘀嘀无限科技发展有限公司 一种数据处理方法及装置
CN111753386A (zh) * 2019-03-11 2020-10-09 北京嘀嘀无限科技发展有限公司 一种数据处理方法及装置
CN111796314A (zh) * 2019-04-09 2020-10-20 Oppo广东移动通信有限公司 信息处理方法、装置、存储介质及电子设备
CN111798260A (zh) * 2019-04-09 2020-10-20 Oppo广东移动通信有限公司 用户行为预测模型构建方法、装置、存储介质及电子设备
CN111161042A (zh) * 2019-11-26 2020-05-15 深圳壹账通智能科技有限公司 个人风险评估方法、装置、终端及存储介质
CN111064617B (zh) * 2019-12-16 2022-07-22 重庆邮电大学 基于经验模态分解聚类的网络流量预测方法及装置
CN111064617A (zh) * 2019-12-16 2020-04-24 重庆邮电大学 基于经验模态分解聚类的网络流量预测方法及装置
CN111738323A (zh) * 2020-06-12 2020-10-02 上海应用技术大学 基于灰色马尔科夫模型的混合增强智能轨迹预测方法和装置
CN111738323B (zh) * 2020-06-12 2024-04-16 上海应用技术大学 基于灰色马尔科夫模型的混合增强智能轨迹预测方法和装置
CN112560910A (zh) * 2020-12-02 2021-03-26 中国联合网络通信集团有限公司 用户分类方法和装置
CN112560910B (zh) * 2020-12-02 2024-03-01 中国联合网络通信集团有限公司 用户分类方法和装置
CN115515143A (zh) * 2021-06-22 2022-12-23 中国移动通信集团重庆有限公司 一种基站覆盖场景确定方法、装置、设备及存储介质
CN113938817A (zh) * 2021-09-10 2022-01-14 哈尔滨工业大学（威海） 一种基于车辆位置信息的车主出行位置预测方法
CN117035176A (zh) * 2023-08-09 2023-11-10 上海智租物联科技有限公司 基于马尔可夫链的用户换电行为预测与定向召回方法
CN117035176B (zh) * 2023-08-09 2024-03-15 上海智租物联科技有限公司 基于马尔可夫链的用户换电行为预测与定向召回方法
CN117370825A (zh) * 2023-10-11 2024-01-09 国网经济技术研究院有限公司 基于注意力条件生成对抗网络的长期场景生成方法及系统
CN117609737A (zh) * 2024-01-18 2024-02-27 中国人民解放军火箭军工程大学 一种惯性导航系统健康状态预测方法、系统、设备及介质
CN117609737B (zh) * 2024-01-18 2024-03-19 中国人民解放军火箭军工程大学 一种惯性导航系统健康状态预测方法、系统、设备及介质

Also Published As

Publication number Publication date
CN106709606B (zh) 2020-10-30
CN106709606A (zh) 2017-05-24

Similar Documents

Publication Publication Date Title
WO2018120428A1 (zh) 个性化场景预测方法、装置、设备和存储介质
WO2018120424A1 (zh) 基于位置服务的人群分类方法、装置、设备和存储介质
WO2018120427A1 (zh) 基于位置服务的风险评估方法、装置、设备和存储介质
US20230409349A1 (en) Systems and methods for proactively providing recommendations to a user of a computing device
WO2018120425A1 (zh) 个人财产状态评估方法、装置、设备和存储介质
WO2018120426A1 (zh) 基于位置服务的个人健康状态评估方法、装置、设备和存储介质
CN103714112B (zh) 定制的事件和景点建议
CN105103185A (zh) 日程偏离通知
US11663282B2 (en) Taxonomy-based system for discovering and annotating geofences from geo-referenced data
CN105283839A (zh) 用以将命令显现在生产力应用用户界面内的个性化社区模型
CN102257493A (zh) 对个人信息进行地理编码
CN103959279A (zh) 基于移动设备生成的事件分析的地图瓦片数据预取
US20140280131A1 (en) Recommendations for Applications Based on Device Context
CN106416318A (zh) 确定与邻近计算设备相关联的数据
CN105122848A (zh) 对环境位置更新进行分组
US9418076B2 (en) System and method for determining interests from location data
CN105103184A (zh) 行程估计
CN105009114B (zh) 预测性地呈现搜索能力
CN105103573A (zh) 图案标记
CN106210006B (zh) 一种设置信息共享方法、移动终端及云服务器
CN110633438B (zh) 一种新闻事件处理的方法、终端、服务器及存储介质
US9084086B2 (en) Matching based on geolocation information from a mobile application
US20200356562A1 (en) Cross-asset data modeling in multi-asset databases
US20170364397A1 (en) Method and system for managing and linking software applications
CN105144220A (zh) 支持计算的基于约束的配置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17886930

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 01.10.2019)

122 Ep: pct application non-entry in european phase

Ref document number: 17886930

Country of ref document: EP

Kind code of ref document: A1