An online sales recommendation system and method based on points of interest

Abstract

The present invention discloses an online sales recommendation system and method based on points of interest. The online sales recommendation method includes establishing a user database, which is used to store the user's account identifier, the user's authentication image, the interest history database, and the shopping database. When a physical store detects that a certain user enters, the face image of the user is compared with the authentication image in the user database. If there is an authentication image in the user database whose similarity with the face image of the user is greater than the similarity threshold, the user is set as a tracked user, and the video image of the tracked user from entering the physical store to leaving the physical store this time is obtained as the analysis image, and the analysis image is analyzed; when a certain user logs in to the online client, the identifier type of the user's account identifier is obtained, and the recommendation method is determined according to the identifier type.

Description

Technical Field

[0001] The present invention relates to the technical field of online sales recommendation, and particularly to an online sales recommendation system and method based on points of interest. Background Art

[0002] With the development of Internet technology, e-commerce has also developed rapidly. More and more merchants adopt a marketing method that combines online and offline, that is, a business model that combines an online store and an offline physical store. This model enables consumers to experience in person in the offline physical store and also purchase online in the online store, thus improving the consumption experience of consumers.

[0003] In the prior art, recommendations for product sales to consumers are made based on the shopping information of consumers online and offline. However, the recommendation methods in the prior art are often relatively single, and the pertinence of product sales recommendations for individual consumers is not strong enough. Summary of the Invention

[0004] The purpose of the present invention is to provide an online sales recommendation system and method based on points of interest to solve the problems raised in the above background art.

[0005] To solve the above technical problems, the present invention provides the following technical solution: An online sales recommendation system based on points of interest, the online sales recommendation system includes a user database, a store detection module, a face comparison module, an image analysis module, and a push judgment module. The user database is used to store the account identification of the user, the authentication image of the user, an interest history database, and a shopping database. The interest history database is used to store the effective interest location information of the user when in the offline store. The shopping database is used to store the shopping information of the user. The store detection module is used to detect whether a user enters an offline store. When it is detected that a certain user enters an offline store, the face comparison module compares the face image of the user with the authentication image in the user database. If there is an authentication image in the user database whose similarity with the face image of the user is greater than a similarity threshold, set the user as a tracked user, and instruct the image analysis module to obtain the video image of the tracked user from entering the offline store to leaving the offline store as an analysis image, and analyze the analysis image. The push judgment module, when a certain user logs in to the online client, obtains the identification type of the account identification of the user, and determines the recommendation method according to the identification type.

[0006] Further, the push judgment module includes an identification acquisition module, a first recommendation module, and a second recommendation module. The identification acquisition module is used to acquire the user's account identification, and determine whether the user's account type is the first identification or the second identification. When the user's account identification is the first identification, the first recommendation module recommends products to the client of the user with reference to the user's shopping database. When the user's account identification is the second identification, the second recommendation module recommends products to the user's client with reference to the user's shopping database and the shopping databases of the user's interested users.

[0007] Further, the image analysis module includes an interest position selection module, an interest ratio calculation module, an interest arrangement module, a boundary ratio selection module, an effective interest position selection module, a variance threshold comparison module, an identification index calculation module, and an identification index comparison module. The interest position selection module determines whether the tracked user in the analyzed image stays at a certain position. If the tracked user in the analyzed image stays at a certain position, then this position is an interest position. The interest ratio calculation module obtains the stay duration ha of the tracked user at a certain interest position. Then, the interest ratio P of this interest position is P = ha / H, where H is the sum of the stay durations of all the interest positions of the tracked user. The interest arrangement module sorts the interest ratios of the respective interest positions of the tracked user in descending order to obtain an interest arrangement. The boundary ratio selection module compares the difference between two adjacent interest ratios in the interest arrangement with an interest difference threshold respectively. When for the first time in the forward direction of the interest arrangement, the difference between two adjacent interest ratios is greater than the interest difference threshold, let the smaller of the two adjacent interest ratios corresponding to the difference be the boundary ratio. The effective interest position selection module selects the interest ratios before the boundary ratio in the interest arrangement as effective interest ratios, and the interest positions corresponding to the effective interest ratios are effective interest positions. The variance threshold comparison module obtains the data information of the tracked user in offline stores in the most recent m times from the interest database, respectively obtains the number of effective interest positions of the tracked user each time in the most recent m times, calculates the variance of the number of effective interest positions in the most recent m times and this time, and compares the variance with a variance threshold. When the variance is less than the variance threshold, the identification index calculation module calculates the identification index of the tracked user this time. where k is the number of effective interest positions this time, h i is the stay duration of the i-th effective interest position, n i is the number of times the i-th effective interest position is an effective interest position in the most recent m times. The identification index comparison module sets the account identification of the tracked user as the first identification when the identification index of the tracked user this time is greater than or equal to an identification threshold, and sets the account identification of the tracked user as the second identification when the variance is greater than or equal to the variance threshold or the identification index of the tracked user this time is less than the identification threshold.

[0008] Further, the online sales recommendation system further includes an interested user selection module, which includes a candidate user selection trigger module, an interval average value acquisition module, and an interval average value comparison module. The candidate user selection trigger module sets a user whose time difference between entering the store after setting a second identifier for the account identifier of the tracked user and the time when the tracked user enters the store is within a preset duration as a candidate user. The interval average value acquisition module acquires the average value of the time intervals between the last m adjacent times of each candidate user entering the offline store. The interval average value comparison module determines that a candidate user is an interested user of the tracked user when the absolute value of the difference between the average value of the time intervals corresponding to a certain candidate user and the average value of the time intervals between the last m adjacent times of the tracked user entering the offline store is less than an absolute reference value.

[0009] An online sales recommendation method based on points of interest, the online sales recommendation method includes the following steps:

[0010] Establish a user database, which is used to store the user's account identifier, the user's authentication image, an interest history database, and a shopping database. The interest history database is used to store the effective interest location information of the user when in the offline store, and the shopping database is used to store the user's shopping information;

[0011] When an offline store detects a user entering, compare the user's face image with the authentication images in the user database. If there is an authentication image in the user database whose similarity to the user's face image is greater than a similarity threshold, set the user as a tracked user, obtain the video image of the tracked user from entering the offline store to leaving the offline store as an analysis image, and analyze the analysis image;

[0012] When a user logs in to the online client, obtain the identifier type of the user's account identifier, and determine the recommendation method according to the identifier type.

[0013] Further, the determining the recommendation method according to the identifier type includes:

[0014] When the user's account identifier is a first identifier, recommend products to the client of the customer with reference to the user's shopping database,

[0015] When the user's account identifier is a second identifier, recommend products to the user's client with reference to the user's shopping database and the shopping databases of the user's interested users.

[0016] Further, the analyzing the analysis image includes the following steps:

[0017] Determine and analyze whether the tracked user stays at a certain location in the image. If the tracked user stays at a certain location in the analyzed image, then this location is an interesting location.

[0018] Obtain the stay duration ha of the tracked user at a certain interesting location. Then, the interest proportion P of this interesting location is P = ha / H, where H is the sum of the stay durations of all interesting locations of the tracked user.

[0019] Sort the interest proportions of each interesting location of the tracked user in descending order to obtain an interest arrangement. Compare the difference between adjacent two interest proportions in the interest arrangement with an interest difference threshold respectively. When comparing for the first time in the forward direction of the interest arrangement and the difference between adjacent two interest proportions is greater than the interest difference threshold, let the smaller one of the two interest proportions corresponding to the difference between adjacent two interest proportions be the demarcation proportion, and let the interest proportion before the demarcation proportion in the interest arrangement be the effective interest proportion, and the interesting location corresponding to the effective interest proportion is the effective interesting location.

[0020] Obtain the data information of the tracked user's most recent m times in offline stores from the interest database. Respectively obtain the number of effective interesting locations of the tracked user each time in the most recent m times, and calculate the variance of the number of effective interesting locations in the most recent m times and this time. Compare the variance with a variance threshold.

[0021] If the variance is less than the variance threshold, calculate the identification index of the tracked user this time. Among them, k is the number of effective interesting locations this time, h i is the stay duration of the i-th effective interesting location. n i is the number of times the i-th effective interesting location is an effective interesting location in the most recent m times.

[0022] If the identification index of the tracked user this time is greater than or equal to an identification threshold, set the account identification of the tracked user to the first identification.

[0023] Otherwise, set the account identification of the tracked user to the second identification.

[0024] Furthermore, after setting the second identification for the tracked user's account identification, it further includes:

[0025] Set the users whose time difference from the time when the tracked user enters the store is within a preset duration as candidate users. Obtain the average value of the time intervals between adjacent two entries into offline stores in the most recent m times for each candidate user.

[0026] If the absolute value of the difference between the average value of the time interval corresponding to a certain candidate user and the average value of the time intervals between two adjacent entries into the offline store among the most recent m times of the tracking user is less than the absolute reference value, then the candidate user is an interest user of the tracking user.

[0027] Compared with the prior art, the beneficial effects achieved by the present invention are as follows: By collecting the video images of users in the offline store, the present invention calculates the interest proportion of users at each interest location, then analyzes the interest proportions of all interest locations to obtain the effective interest locations, compares and analyzes the interest locations of this time with the historical interest locations, thereby analyzing the type of users, and then selects a suitable commodity sales recommendation method according to the type of users, so that the recommendation method is more specific and targeted, and the online shopping experience of users is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0029] Figure 1 is a schematic structural diagram of the online sales recommendation system and method based on points of interest of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0031] Please refer to Figure 1, the present invention provides a technical solution: an online sales recommendation system based on points of interest. The online sales recommendation system includes a user database, a store detection module, a face comparison module, an image analysis module, and a push judgment module. The user database is used to store the user's account identifier, the user's authentication image, an interest history database, and a shopping database. The interest history database is used to store the effective interest location information of the user when in a physical store. The shopping database is used to store the user's shopping information. The store detection module is used to detect whether a user enters a physical store. When it detects that a certain user enters a physical store, the face comparison module compares the face image of the user with the authentication image in the user database. If there is an authentication image in the user database whose similarity with the face image of the user is greater than a similarity threshold, set this user as a tracked user, and instruct the image analysis module to obtain the video image of the tracked user from entering the physical store to leaving the physical store as the analysis image, and analyze the analysis image. The push judgment module, when a certain user logs in to the online client, obtains the identifier type of the user's account identifier, and determines the recommendation method according to the identifier type.

[0032] The push judgment module includes an identifier acquisition module, a first recommendation module, and a second recommendation module. The identifier acquisition module is used to obtain the user's account identifier and determine whether the user's account type is a first identifier or a second identifier. When the user's account identifier is the first identifier, the first recommendation module recommends products to the client of this customer with reference to the user's shopping database. When the user's account identifier is the second identifier, the second recommendation module recommends products to the client of this user with reference to the user's shopping database and the shopping database of the user's interest users.

[0033] The image analysis module includes an interest position selection module, an interest ratio calculation module, an interest arrangement module, a boundary ratio selection module, a valid interest position selection module, a variance threshold comparison module, an identification index calculation module, and an identification index comparison module. The interest position selection module determines whether the tracked user stays at a certain position in the analyzed image. If the tracked user stays at a certain position in the analyzed image, then this position is an interest position. The interest ratio calculation module obtains the stay duration ha of the tracked user at a certain interest position. Then, the interest ratio P of this interest position is P = ha / H, where H is the sum of the stay durations of all the interest positions of the tracked user. The interest arrangement module sorts the interest ratios of each interest position of the tracked user in descending order to obtain an interest arrangement. The boundary ratio selection module compares the difference between two adjacent interest ratios in the interest arrangement with an interest difference threshold respectively. When, in the forward direction of the interest arrangement, it is first compared that the difference between two adjacent interest ratios is greater than the interest difference threshold, let the smaller one of the two adjacent interest ratios corresponding to the difference be the boundary ratio. The valid interest position selection module selects the interest ratios before the boundary ratio in the interest arrangement as valid interest ratios, and the interest positions corresponding to the valid interest ratios are valid interest positions. The variance threshold comparison module obtains the data information of the tracked user's last m times in the offline store from the interest database, respectively obtains the number of valid interest positions of the tracked user each time in the last m times, calculates the variance of the number of valid interest positions in the last m times and this time, and compares the variance with the variance threshold. When the variance is less than the variance threshold, the identification index calculation module calculates the identification index of the tracked user this time. where k is the number of valid interest positions this time, and hi is the stay duration of the i-th valid interest position. ni is the number of times that the i-th valid interest position is a valid interest position in the last m times. When the identification index of the tracked user this time is greater than or equal to the identification threshold, the identification index comparison module sets the account identification of the tracked user to the first identification. When the variance is greater than or equal to the variance threshold or the identification index of the tracked user this time is less than the identification threshold, the identification index comparison module sets the account identification of the tracked user to the second identification.

[0034] The online sales recommendation system further includes an interested user selection module, which includes a candidate user selection trigger module, an interval average value acquisition module, and an interval average value comparison module. The candidate user selection trigger module sets a user whose time difference between entering the store after setting a second identifier for the account identifier of the tracked user and the time when the tracked user enters the store is within a preset duration as a candidate user. The interval average value acquisition module acquires the average value of the time intervals between the last m adjacent times of each candidate user entering the offline store. The interval average value comparison module makes a candidate user an interested user of the tracked user when the absolute value of the difference between the average value of the time intervals corresponding to a certain candidate user and the average value of the time intervals between the last m adjacent times of the tracked user entering the offline store is less than an absolute reference value.

[0035] An online sales recommendation method based on points of interest, the online sales recommendation method includes the following steps:

[0036] Establish a user database, which is used to store the user's account identifier, the user's authentication image, an interest history database, and a shopping database. The interest history database is used to store the effective interest location information of the user each time when in the offline store. The shopping database is used to store the shopping information of the user; the user's authentication image is the face image of the corresponding user, and the shopping database stores both the user's online shopping information and the user's offline shopping information;

[0037] When an offline store detects that a certain user enters, compare the face image of the user with the authentication images in the user database. If there is an authentication image in the user database whose similarity with the face image of the user is greater than a similarity threshold, set this user as a tracked user, obtain the video image of the tracked user from entering the offline store to leaving the offline store as an analysis image, and analyze the analysis image;

[0038] The analysis of the analysis image includes the following steps:

[0039] Judge whether the tracked user in the analysis image stays at a certain position. If the tracked user in the analysis image stays at a certain position, then this position is an interest position.

[0040] Obtain the stay duration ha of the tracked user at a certain interest position. Then the interest proportion P of this interest position is P = ha / H, where H is the sum of the stay durations of all the interest positions of the tracked user.

[0041] Sort the interest ratios of each interest location tracking the user from largest to smallest to obtain an interest arrangement. Compare the differences between adjacent interest ratios in the interest arrangement with the interest difference threshold one by one. When the difference between adjacent interest ratios is first compared to be greater than the interest difference threshold in the forward direction of the interest arrangement, let the smaller of the two interest ratios corresponding to the difference between adjacent interest ratios be the demarcation ratio, and let the interest ratios before the demarcation ratio in the interest arrangement be the effective interest ratios, and the interest locations corresponding to the effective interest ratios be the effective interest locations. For example, in this application, the interest arrangement is 0.24, 0.21, 0.20, 0.17, 0.1, 0.07, 0.01, and the interest difference threshold is 0.05. Then the differences between adjacent interest ratios are 0.03, 0.01, 0.03, 0.07, 0.03, 0.06. So in this embodiment, 0.1 is the demarcation ratio, and the interest locations corresponding to 0.24, 0.21, 0.20, 0.17 are the effective interest locations;

[0042] When tracking that the user stays at a certain location, it indicates that the tracked user may be interested in the products at that location. However, it is possible that the tracked user is not really interested. Therefore, further judgment is made by tracking the interest ratios of the user at the interest locations. When the difference between adjacent interest ratios is less than or equal to the interest difference threshold, it means that the interests in the two corresponding interest locations are about the same. When the difference between adjacent interest ratios is greater than the interest difference threshold, it means that the tracked user is more interested in the previous interest location. Relatively speaking, the tracked user is less interested in the latter interest location. Then it means that the tracked user is really interested in the previous interest location and not really interested in the latter interest location. In this application, the effective interest locations are obtained by comparing the interest arrangement, making the analysis result of this application more accurate.

[0043] Obtain the data information of the tracked user's most recent m times in the offline store from the interest database. Respectively obtain the number of effective interest locations of the tracked user each time in the most recent m times, and calculate the variance of the number of effective interest locations of the most recent m times and this time. Compare the variance with the variance threshold. For example, m = 4, the number of effective interest locations in the most recent m times is 5, 4, 5, 6, and the number of effective interest locations this time is 5. Then calculate the variance of 5, 4, 5, 6, 5, which is the variance of the number of effective interest locations of the most recent m times and this time;

[0044] If the variance is less than the variance threshold, calculate the identification index of the tracked user this time where k is the number of effective interest locations this time, hi is the residence duration of the i-th effective interest location, and H is the sum of the residence durations of the k effective interest locations. ni is the number of times that the i-th effective interest location is an effective interest location in the most recent m times;

[0045] If the identification index of the tracked user for this time is greater than or equal to the identification threshold, set the account identification of the tracked user to the first identification.

[0046] If the variance is greater than or equal to the variance threshold or the identification index of the tracked user for this time is less than the identification threshold, set the account identification of the tracked user to the second identification; in this application, judgment is made every time a user enters an offline store, and the corresponding account identification may be adjusted, so that the recommendation mode of this application can be adaptively adjusted to improve the accuracy of the recommendation result.

[0047] After setting the second identification for the account identification of the tracked user, it further includes:

[0048] Set the users whose time difference from the time when the tracked user enters the store is within the preset time as candidate users, and obtain the average value of the time intervals between two adjacent entries into the offline store in the most recent m times for each candidate user.

[0049] If the absolute value of the difference between the average value of the time intervals corresponding to a certain candidate user and the average value of the time intervals between two adjacent entries into the offline store in the most recent m times of the tracked user is less than the absolute reference value, then this candidate user is an interest user of this tracked user.

[0050] When a certain user logs in to the online client, obtain the identification type of the account identification of this user, and determine the recommendation method according to the identification type; the determining the recommendation method according to the identification type includes:

[0051] When the account identification of this user is the first identification, recommend products to the client of this customer with reference to the shopping database of this user.

[0052] When the account identification of this user is the second identification, recommend products to the client of this user with reference to the shopping database of this user and the shopping database of the interest users of this user. When shopping, some people like to try new things and like to receive and try new things, while some people's preferences are relatively fixed and they are only interested in specific things. When the variance is less than the variance threshold, it means that the number of effective interest positions of this user is almost the same each time, and it may belong to people with relatively fixed preferences. The identification index of the user is used to further judge whether this user belongs to people with relatively fixed preferences. The larger the identification index, the greater the similarity between the effective interest position of this user for this time and the effective interest positions when entering the store before, indicating that the preferences of this user are relatively fixed. Therefore, recommend products to the user only according to the historical shopping information of the user; when the variance is greater than or equal to the variance threshold or the identification index of the tracked user for this time is less than the identification threshold, it means that the preferences of this user are relatively extensive and unrestricted. Therefore, recommend products to the user according to the historical shopping information of this user and the interest users of this user.

[0053] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0054] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An online sales recommendation system based on points of interest, characterized in that: The online sales recommendation system includes a user database, a store detection module, a face comparison module, an image analysis module, and a push judgment module. The user database is used to store the user's account identifier, the user's authentication image, an interest history database, and a shopping database. The interest history database is used to store the effective interest location information of the user when in a physical store. The shopping database is used to store the user's shopping information. The store detection module is used to detect whether a user enters a physical store. When it detects that a certain user enters a physical store, the face comparison module compares the user's face image with the authentication image in the user database. If there is an authentication image in the user database whose similarity with the user's face image is greater than the similarity threshold, set this user as a tracked user, and let the image analysis module obtain the video image of the tracked user from entering the physical store to leaving the physical store as the analysis image, and analyze the analysis image. The push judgment module, when a certain user logs in to the online client, obtains the identifier type of the user's account identifier, and determines the recommendation method according to the identifier type; The push judgment module includes an identifier acquisition module, a first recommendation module, and a second recommendation module. The identifier acquisition module is used to obtain the user's account identifier and determine whether the user's account type is a first identifier or a second identifier. When the user's account identifier is the first identifier, the first recommendation module recommends products to the client of this customer with reference to the user's shopping database. When the user's account identifier is the second identifier, the second recommendation module recommends products to the user's client with reference to the user's shopping database and the shopping database of the user's interest users; The image analysis module includes an interest position selection module, an interest ratio calculation module, an interest arrangement module, a boundary ratio selection module, a valid interest position selection module, a variance threshold comparison module, an identification index calculation module, and an identification index comparison module. The interest position selection module determines whether the tracked user stays at a certain position in the analyzed image. If the tracked user stays at a certain position in the analyzed image, then this position is an interest position. The interest ratio calculation module obtains the stay duration ha of the tracked user at a certain interest position. Then, the interest ratio P of this interest position is P = ha / H, where H is the sum of the stay durations of all interest positions of the tracked user. The interest arrangement module sorts the interest ratios of each interest position of the tracked user in descending order to obtain an interest arrangement. The boundary ratio selection module compares the difference between two adjacent interest ratios in the interest arrangement with the interest difference threshold respectively. When the difference between two adjacent interest ratios is first compared to be greater than the interest difference threshold in the forward direction of the interest arrangement, let the smaller one of the two adjacent interest ratios corresponding to the difference be the boundary ratio. The valid interest position selection module selects the interest ratios before the boundary ratio in the interest arrangement as valid interest ratios, and the interest positions corresponding to the valid interest ratios are valid interest positions. The variance threshold comparison module obtains the data information of the tracked user's last m times in the offline store from the interest database, respectively obtains the number of valid interest positions of the tracked user each time in the last m times, calculates the variance of the number of valid interest positions in the last m times and this time, and compares the variance with the variance threshold. When the variance is less than the variance threshold, the identification index calculation module calculates the identification index of the tracked user this time. , where, is the number of valid interest positions this time, is the stay duration of the i-th valid interest position, , is the number of times the i-th valid interest position is a valid interest position in the last m times. When the identification index of the tracked user this time is greater than or equal to the identification threshold, the identification index comparison module sets the account identification of the tracked user to the first identification. When the variance is greater than or equal to the variance threshold or the identification index of the tracked user this time is less than the identification threshold, the identification index comparison module sets the account identification of the tracked user to the second identification.

2. The online sales recommendation system based on points of interest according to claim 1, characterized in that: The online sales recommendation system further includes an interest user selection module. The interest user selection module includes a candidate user selection trigger module, an interval average value acquisition module, and an interval average value comparison module. The candidate user selection trigger module sets, after setting the second identifier for the tracked user's account identifier, the users whose time difference from the time when the tracked user enters the store is within the preset time period as candidate users. The interval average value acquisition module obtains the average value of the time intervals between the last m adjacent times of each candidate user entering the physical store. The interval average value comparison module, when the absolute value of the difference between the average value of the time intervals corresponding to a certain candidate user and the average value of the time intervals between the last m adjacent times of the tracked user entering the physical store is less than the absolute reference value, sets this candidate user as the interest user of this tracked user.

3. An online sales recommendation method based on points of interest, used to implement the online sales recommendation system based on points of interest according to claim 1, characterized in that: The online sales recommendation method includes the following steps: Establish a user database, which is used to store the user's account identifier, the user's authentication image, an interest history database, and a shopping database. The interest history database is used to store the effective interest location information of the user when in a physical store. The shopping database is used to store the user's shopping information; When an offline store detects the entry of a certain user, it compares the face image of the user with the authentication images in the user database. If there is an authentication image in the user database whose similarity with the face image of the user is greater than the similarity threshold, set this user as the tracked user, and obtain the video image of the tracked user from entering the offline store to leaving the offline store as the analysis image, and analyze the analysis image; When a certain user logs in to the online client, obtain the identification type of the account identification of the user, and determine the recommendation method according to the identification type.

4. The online sales recommendation method based on points of interest according to claim 3, characterized in that: The determining the recommendation method according to the identification type includes: When the account identification of the user is the first identification, recommend products to the client of the customer with reference to the shopping database of the user. When the account identification of the user is the second identification, recommend products to the client of the user with reference to the shopping database of the user and the shopping databases of the user's interest users.

5. The online sales recommendation method based on points of interest according to claim 4, characterized in that: The analyzing the analysis image includes the following steps: Judge whether the tracked user in the analysis image stays at a certain position. If the tracked user in the analysis image stays at a certain position, then this position is the interest position. Obtain the stay duration ha of the tracked user at a certain interest position. Then the interest proportion P of this interest position is P = ha / H, where H is the sum of the stay durations of all the interest positions of the tracked user. Sort the interest proportions of each interest position of the tracked user from largest to smallest to obtain an interest arrangement. Compare the difference between two adjacent interest proportions in the interest arrangement with the interest difference threshold respectively. When the difference between two adjacent interest proportions is greater than the interest difference threshold for the first time in the forward direction along the interest arrangement, set the smaller of the two adjacent interest proportions corresponding to the difference as the demarcation proportion, and set the interest proportion before the demarcation proportion in the interest arrangement as the effective interest proportion, and the interest position corresponding to the effective interest proportion as the effective interest position. Obtain the data information of the tracked user in the offline store in the most recent m times from the interest database, obtain the number of effective interest positions of the tracked user each time in the most recent m times respectively, calculate the variance of the number of effective interest positions in the most recent m times and this time, and compare the variance with the variance threshold. If the variance is less than the variance threshold, calculate the identification index of the user for this time , where is the number of valid interest positions for this time, is the residence duration of the i-th valid interest position, , is the number of times the i-th valid interest position has been a valid interest position in the most recent m times; If the identification index of the tracked user this time is greater than or equal to the identification threshold, set the account identification of the tracked user as the first identification. Otherwise, set the account identification of the tracked user as the second identification.

6. The online sales recommendation method based on points of interest according to claim 5, characterized in that: After setting the second identification for the account identification of the tracked user, it further includes: Set the users whose time difference from the time when the tracked user enters the store is within the preset duration as candidate users, and obtain the average value of the time intervals between two adjacent entries into the offline store in the most recent m times for each candidate user. If the absolute value of the difference between the average value of the time intervals corresponding to a certain candidate user and the average value of the time intervals between two adjacent entries into the offline store in the most recent m times of the tracked user is less than the absolute reference value, then this candidate user is an interest user of this tracked user.

Images