A multi-source data real-time acquisition and fusion method and system

Abstract

The present application relates to the field of data processing, and more particularly to a multi-source data real-time collection and fusion method and system, the system comprising: a portrait analysis unit configured to determine whether to perform portrait optimization according to a portrait coverage rate and an inefficient label proportion; a portrait optimization unit configured to determine a corresponding image optimization mode according to an operation frequency and a demand stability coefficient of a target user, the corresponding image optimization mode being either optimization processing according to an effective demand difference or optimization processing according to a reference data state; an image analysis unit configured to determine a portrait supplement parameter according to an image diversity of a reference source and a target image proportion, the portrait supplement parameter being either an effective mean value or an image matching coefficient mean value; and a text analysis unit configured to determine a text structure coefficient according to a keyword diversity of an effective keyword and a keyword distribution density, and to determine a portrait supplement parameter according to a difference degree of a corresponding text structure coefficient of a reference source; the present application effectively improves the accuracy of a target user portrait.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to a multi-source data real-time collection and fusion method and system. BACKGROUND

[0002] With the rapid development of the e-commerce industry, the online shopping data platform plays an important role in precise marketing, user experience optimization and other fields due to its massive user behavior and personalized recommendation capabilities. Among them, user portrait construction is the core technology to ensure that the platform achieves precise touch in diversified consumption scenarios. The accuracy of the portrait directly affects the conversion rate of the recommendation system and user satisfaction. However, the existing user portrait technology often fails to effectively capture dynamic user preferences and optimize user portraits in a timely manner when dealing with changing consumption behaviors, resulting in poor applicability of user portraits and a decline in recommendation accuracy. Therefore, how to accurately portray and update user portraits in a complex and changing online shopping environment is a key technical problem that needs to be solved by those skilled in the art.

[0003] Chinese Patent Publication No. CN116049553A discloses a user portrait construction method and system based on multi-source information, which includes: classifying shopping information according to product information to obtain several shopping categories; naming each portrait category that needs to be constructed, the name of the portrait category that needs to be constructed is a person or an object, obtaining an object portrait category and one or more person portrait categories; in addition, the present application generates account binding information for each person portrait category, receives user input of related accounts, and binds the person portrait category with the related accounts; retrieve the shopping information of the related accounts bound by the person portrait category, and construct the person portrait for the person portrait category according to the retrieved shopping information and the shopping information of the corresponding person portrait category, to ensure more accurate construction of the person portrait. It can be seen that the above technical solution has the following problems: although the influence of information sources on the accuracy of user portrait construction is considered, the need for continuous optimization and adjustment of user portraits in actual scenarios is not considered, and the influence of multi-source information data on portrait accuracy during the optimization of user portraits, resulting in poor optimization effect of user portraits. SUMMARY

[0004] Therefore, the present application provides a multi-source data real-time collection and fusion method and system to overcome the problem that the prior art does not consider the need for continuous optimization and adjustment of user portraits in actual scenarios, the influence of multi-source information data on portrait accuracy during the optimization of user portraits, and the poor optimization effect of user portraits.

[0005] To achieve the above-mentioned purpose, the present application provides a multi-source data real-time collection and fusion system, comprising:

[0006] The image analysis unit is configured to determine the accuracy state of the image according to the image coverage rate and the proportion of the ineffective label, and determine that the image optimization is performed when the image coverage rate is less than or equal to a preset image coverage rate and the proportion of the ineffective label is greater than a preset proportion of the ineffective label.

[0007] The image optimization unit is connected with the image analysis unit and configured to determine the operation state of the target user according to the operation frequency and the demand stability coefficient of the target user, and determine the corresponding image optimization mode according to the operation state of the target user, that is, the optimization processing is performed according to the effective demand difference or the optimization processing is performed according to the reference data state. The image supplement mode determined according to the effective demand difference is the associated category supplement or the related user supplement, and the optimization auxiliary mode determined according to the reference data state is the image optimization or the keyword optimization.

[0008] The image analysis unit is connected with the image analysis unit and the image optimization unit, and configured to determine the target image difference degree of the reference source according to the image diversity of the reference source and the target image proportion, and determine the image supplement parameter according to the target image difference degree, that is, the effective mean or the image matching coefficient mean.

[0009] The text analysis unit is connected with the image analysis unit, the image optimization unit and the image analysis unit, and configured to determine the text structure coefficient according to the diversity and the distribution density of the effective keyword, and determine the image supplement parameter according to the difference degree of the text structure coefficient of the reference source.

[0010] Further, the image analysis unit periodically determines the accuracy state of the image of each target user according to the image coverage rate and the proportion of the ineffective label, and determines that the image optimization is performed when the image coverage rate is less than or equal to a preset image coverage rate and the proportion of the ineffective label is greater than a preset proportion of the ineffective label.

[0011] Further, the image optimization unit determines the operation state of the target user according to the operation frequency and the demand stability coefficient of the target user in response to a preset optimization condition, and determines the corresponding image optimization mode according to the operation state of the target user.

[0012] If the operation frequency of the target user is less than a preset operation frequency and the demand stability coefficient is greater than or equal to a preset demand stability coefficient, the image optimization mode is the optimization processing according to the effective demand difference.

[0013] If the operation frequency of the target user is greater than or equal to a preset operation frequency or the demand stability coefficient is less than a preset demand stability coefficient, the image optimization mode is the optimization processing according to the reference data state.

[0014] The preset optimization condition is that the image analysis unit determines that the image optimization is performed.

[0015] Furthermore, the portrait optimization unit responds to the first optimization processing condition and determines the portrait supplementation method based on the comparison result between the effective demand difference and the preset effective demand difference;

[0016] If the effective demand difference is greater than the preset effective demand difference, the profile will be supplemented based on related product categories.

[0017] If the effective demand difference is less than or equal to the preset effective demand difference, the profile will be supplemented based on the relevant users.

[0018] The first optimization condition is to determine the portrait optimization method as being based on the difference in effective demand.

[0019] Furthermore, the image optimization unit responds to the second optimization processing condition, determines the reference data status based on the image and text stability coefficient and the bias stability coefficient corresponding to the reference source, and determines the optimization assistance method based on the reference data status.

[0020] If the reference data state shows that the image-text stability coefficient is greater than the preset image-text stability coefficient and the bias stability coefficient is greater than the preset bias stability coefficient, then the optimization assistance method is image optimization.

[0021] If the reference data shows that the stability coefficient of the image and text is greater than the preset stability coefficient of the image and text and the emphasis stability coefficient is less than or equal to the preset emphasis stability coefficient, then the optimization assistance method is keyword optimization.

[0022] The second optimization condition is to determine the portrait optimization method as being based on the status of the reference data.

[0023] Furthermore, in response to the first optimization assistance mode condition, the image analysis unit determines the degree of difference between the target image and the reference source based on the image diversity of the reference source and the proportion of the target image, and determines the image supplementation parameters based on the degree of difference between the target image.

[0024] If the difference in the target image is less than the preset difference in the target image, then the effective mean of the reference source is extracted and recorded as the image supplementation parameter for image supplementation.

[0025] If the difference in the target image is greater than or equal to the preset difference in the target image, the image supplementation parameters are determined based on the image matching coefficient.

[0026] The first optimization assistance method condition is that the optimization assistance method is confirmed to be image optimization.

[0027] Furthermore, the image analysis unit responds to the preset supplementary parameter conditions and determines whether to supplement parameters based on the comparison result between the image matching coefficient and the preset image matching coefficient;

[0028] If the image matching coefficient is greater than the preset image matching coefficient, the average image matching coefficient will be used as a supplementary parameter for image supplementation.

[0029] If the image matching coefficient is less than or equal to the preset image matching coefficient, the image supplementation will fail.

[0030] The preset supplementary parameter condition is to determine the portrait supplementary parameters based on the image matching coefficient.

[0031] Furthermore, in response to the second optimization assistance condition, the text analysis unit determines the region category based on the dwell time and browsing speed, and records the region corresponding to the browsing speed being less than the preset browsing speed and the dwell time being greater than the preset dwell time as the quick lookup dwell area, and records the text in the quick lookup dwell area as effective keywords;

[0032] The second optimization assistance method is determined by confirming that the optimization assistance method is keyword optimization.

[0033] Furthermore, the text analysis unit determines the text structure coefficient based on the keyword diversity and keyword distribution density of effective keywords, and determines the profile supplementary parameters based on the difference in the text structure coefficients corresponding to the reference source.

[0034] If the difference in the text structure coefficient is less than or equal to the preset difference, then the supplementary parameter for the portrait is the text structure coefficient;

[0035] If the difference in the text structure coefficients is greater than the preset difference, the portrait supplementation will fail.

[0036] The present invention also provides a method for applying the real-time acquisition and fusion system of the multi-source data, comprising:

[0037] Periodically determine the accuracy status of the profiles for each target user based on profile coverage and the proportion of inefficient tags, and decide whether to optimize the profiles based on the accuracy status.

[0038] The corresponding profile optimization method is determined based on the target user's operation frequency and demand stability coefficient: optimization based on the effective demand difference or optimization based on the status of reference data.

[0039] When optimizing based on the effective demand difference, the method of supplementing the profile is determined based on the comparison result between the effective demand difference and the preset effective demand difference.

[0040] When performing optimization based on the status of reference data, the optimization assistance method is determined to be either image optimization or keyword optimization based on the image and text stability coefficient and the bias stability coefficient corresponding to the reference source.

[0041] During image optimization, the degree of difference between the reference source and the target image is determined based on the image diversity of the reference source and the proportion of the target image. The portrait supplementation parameters are then determined to be either the effective mean or the mean of the image matching coefficients based on the degree of difference of the target image.

[0042] When optimizing keywords, the text structure coefficient is determined based on the keyword diversity and keyword distribution density of effective keywords, and the difference between the text structure coefficients of the reference source and the text structure coefficients is used to determine whether the supplementary parameters of the profile are the difference of the text structure coefficients.

[0043] Compared with the prior art, the beneficial effect of the present invention is that the technical solution of the present invention periodically detects the accuracy status of the target user's profile and determines whether the profile needs to be optimized based on the accuracy status. The applicability of the current profile is reflected by the profile coverage and the proportion of inefficient tags. This avoids the poor effectiveness of the obtained profile accuracy status caused by relying on only a single influencing factor to determine whether the profile needs to be optimized in the process of profile optimization in the prior art, thereby improving the accuracy of subsequent profile optimization.

[0044] Furthermore, in this invention, the target user's operation status is determined based on the user's operation frequency and demand stability coefficient. The operation frequency and demand stability coefficient reflect the user's shopping activity and preference stability. Different profile optimization methods are determined based on the user's operation status. Specifically, when optimizing based on the effective demand difference, the initial user profile data is limited, so the user profile is supplemented, with different supplementation methods corresponding to the effective demand difference. When optimizing based on reference data, the initial user profile data is richer, so the user profile is optimized with assistance, and different optimization assistance methods are determined based on the reference data status. This avoids the problem in existing technologies where a single profile optimization method cannot meet the profile optimization needs in real-world scenarios, resulting in poor profile optimization effects. This invention determines different optimization methods based on the initial user profile data and adds auxiliary learning parameters during optimization assistance, thereby improving the accuracy of profile optimization.

[0045] Furthermore, in the technical solution of this invention, the reference data state is determined based on the image-text stability coefficient and the bias stability coefficient corresponding to the reference source. The image-text stability coefficient and the bias stability coefficient reflect the target user's sensitivity to images and text, and corresponding image optimization or keyword optimization is performed. During the image optimization process, the usability and effectiveness of the target image are reflected by comparing the degree of difference between the target images of the reference sources, and different profile supplementation parameters are determined according to the degree of difference. When the profile supplementation parameter is an image matching coefficient, profile supplementation is only performed on images with high matching degrees, avoiding the problem that a single profile supplementation parameter cannot meet the profile optimization needs corresponding to the actual reference source, resulting in poor rationality of the profile supplementation parameter, thereby enhancing the profile optimization effect.

[0046] Furthermore, in this invention, the quick-search dwell area is determined based on dwell time and browsing speed. Dwell time and browsing speed reflect the user's level of attention and interaction depth. Effective keywords are identified based on user behavior characteristics, improving the accuracy of effective keywords. The text structure coefficient is determined based on the diversity and distribution density of effective keywords, reflecting the information organization method of the text structure. For text structure coefficients with small differences, the text structure coefficient is used as a supplementary parameter for profile supplementation. The difference reflects the effectiveness of using the text structure coefficient, avoiding indiscriminate use of the text structure coefficient for profile supplementation, which leads to unsatisfactory supplementation results, reduced text analysis efficiency, and thus enhanced profile optimization effect. Attached Figure Description

[0047] Figure 1 This is a unit connection diagram of the real-time acquisition and fusion system for multi-source data of the present invention;

[0048] Figure 2 This is a flowchart illustrating how the present invention determines the corresponding profile optimization method based on the target user's operation frequency and demand stability coefficient.

[0049] Figure 3 This is a flowchart illustrating how the present invention determines the method for supplementing a profile based on the difference in effective demand.

[0050] Figure 4 This is a schematic diagram of the method for real-time acquisition and fusion of multi-source data according to the present invention. Detailed Implementation

[0051] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0052] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0053] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0054] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0055] Please see Figures 1 to 3 As shown, the present invention provides a real-time acquisition and fusion system for multi-source data, comprising:

[0056] The portrait analysis unit is used to determine the accuracy status of the portrait based on the portrait coverage rate and the proportion of inefficient tags. When the portrait accuracy status is less than or equal to the preset portrait coverage rate and the proportion of inefficient tags is greater than the preset inefficient tag proportion, it is determined that portrait optimization should be performed.

[0057] The profile optimization unit, connected to the profile analysis unit, is used to determine the target user's operation status based on the target user's operation frequency and demand stability coefficient, and to determine the corresponding image optimization method based on the target user's operation status, either by optimizing based on the effective demand difference or by optimizing based on the reference data status; wherein, the profile supplementation method determined based on the effective demand difference is related category supplementation or related user supplementation; and the optimization assistance method determined based on the reference data status is image optimization or keyword optimization.

[0058] An image analysis unit, which is connected to the image analysis unit and the image optimization unit respectively, is used to determine the degree of difference between the target image corresponding to the reference source based on the image diversity of the reference source and the proportion of the target image, and to determine the image supplementation parameters as the effective mean or the mean of the image matching coefficient based on the degree of difference of the target image.

[0059] The text analysis unit, which is connected to the portrait analysis unit, the portrait optimization unit, and the image analysis unit, is used to determine the text structure coefficient based on the keyword diversity and keyword distribution density of effective keywords, and to determine the portrait supplementary parameters based on the difference in the text structure coefficients corresponding to the reference source.

[0060] The application scenario of this invention is to optimize user profiles for online shopping platforms. This invention sets a continuous cyclical monitoring cycle, wherein the duration of a single monitoring cycle can be adaptively set by the user according to actual application needs. This invention provides a value for the duration of a single monitoring cycle, with the duration of a single monitoring cycle being 1 month. The target platform is an online shopping platform for user profile optimization.

[0061] This invention utilizes several historical records. Each historical record includes at least the following metrics: image coverage, proportion of inefficient tags, browsing duration, operation frequency, demand stability coefficient, effective demand difference, weight coefficient, image and text stability coefficient, bias stability coefficient, target image difference degree, image matching coefficient, browsing speed, dwell time, text structure coefficient, and difference degree. Each historical record is also assigned a qualification marker, which indicates whether the historical record meets user needs. To determine whether a historical record meets user needs, a questionnaire can be used to obtain usage ratings from platform users regarding purchase recommendations. It is understood that determining whether a historical record meets user needs based on self-defined indicators is a skill already known to those skilled in the art and is not limited here.

[0062] Specifically, the profile analysis unit periodically determines the profile accuracy status of each target user based on profile coverage and the proportion of inefficient tags. When the profile accuracy status is that the profile coverage is less than or equal to the preset profile coverage and the proportion of inefficient tags is greater than the preset proportion of inefficient tags, it determines that profile optimization should be performed.

[0063] In particular, when the portrait accuracy status is defined as the portrait coverage rate being greater than or equal to the preset portrait coverage rate or the proportion of inefficient tags being less than or equal to the preset proportion of inefficient tags, the portrait is considered qualified and no portrait optimization is required.

[0064] The target users are those who have registered on the online shopping platform. A user profile tag set is set for each target user. The user profile tag set records several profile tags. Among them, profile tags are generated by analyzing historical data such as users' historical purchasing behavior through machine learning. Each purchased product is also set with a corresponding product category tag.

[0065] Extract all products purchased by the target user in the most recent monitoring period, and extract the product category tags corresponding to all purchased products. Then, the profile coverage rate for this monitoring period = the number of common tags between the product category tags and the user profile tag set / the number of product category tags purchased. For example, if a product category tag is cosmetics, and a profile tag in the user profile tag set is also cosmetics, then it is recorded as a common tag. Detect the profile coverage rate for several monitoring periods and record the average value as the user profile coverage rate.

[0066] Inefficient tag percentage = number of inefficient tags in the target user's tag set / number of tags in the target user's tag set; detect the number of purchases corresponding to each tag in the target user's tag set, and tag whose number of purchases corresponding to each tag is less than the preset number of purchases is recorded as an inefficient tag.

[0067] The preset purchase count can be set adaptively by the user according to the actual application scenario. It is understood that the greater the user's tolerance for the impact of inefficient tags, the smaller the preset purchase count will be. This invention provides a method for determining the preset purchase count, which extracts the corresponding purchase count from the historical records that meet the user's needs, filters out outliers, and records the average of the purchase counts after removing outliers as the preset purchase count.

[0068] Users can adaptively set the preset image coverage rate and preset inefficient label ratio according to their actual application needs. It is understood that the greater the user's need for image optimization, the smaller the preset image coverage rate and the smaller the preset inefficient label ratio. This invention provides a method for setting the preset image coverage rate and preset inefficient label ratio. The method extracts the corresponding image coverage rate and inefficient label ratio from the historical records that meet the user's needs, filters out outliers, and records the average values ​​of the image coverage rate and inefficient label ratio after removing outliers as the preset image coverage rate and preset inefficient label ratio, respectively. The outlier filtering method can be, but is not limited to, the 3σ criterion method or the IQR method.

[0069] Specifically, the profile optimization unit responds to preset optimization conditions, determines the user's operation status based on the target user's operation frequency and demand stability coefficient, and determines the corresponding profile optimization method based on the user's operation status.

[0070] If the user's operation status is that the operation frequency is less than the preset operation frequency and the demand stability coefficient is greater than or equal to the preset demand stability coefficient, then the profile optimization method is to optimize based on the effective demand difference.

[0071] If the user's operation status is that the operation frequency is greater than or equal to the preset operation frequency or the demand stability coefficient is less than the preset demand stability coefficient, then the profile optimization method is to optimize based on the reference data status.

[0072] The preset optimization condition is determined by the portrait analysis unit to optimize the portrait.

[0073] The operation frequency is the number of valid operations for the target user within the most recent monitoring period.

[0074] Extract the target user's browsing history for the most recent monitoring period. The browsing history includes the user's online time, the categories of goods purchased, and the frequency of purchases. The number of valid browsing records is counted as the valid count. The browsing history is a complete process from when the user logs into the target platform to when they leave the target platform. Valid browsing records are those where the online time on the target platform is longer than the preset online time.

[0075] The preset online duration value can be adaptively set by the user according to actual application needs. This invention provides a method for setting the preset online duration value, which extracts the minimum online duration value corresponding to the historical records that meet the user's needs, filters out outliers, and records the average of the minimum online duration after removing outliers as the preset online duration. This invention provides a preset online duration value, in which the preset online duration is 5 minutes.

[0076] For a single monitoring period, the corresponding demand stability coefficient = ;in, H represents the number of times the target user purchases the product tag i corresponding to the product, H represents the total number of purchases by the target user, and n represents the total number of product tags.

[0077] The preset operation frequency and preset demand stability coefficient can be adaptively set by the user according to actual application needs. This invention provides a method for determining the preset operation frequency and preset demand stability coefficient. The method extracts the corresponding operation frequency and demand stability coefficient from the historical records that meet the user's needs, filters out outliers, and records the average values ​​of the operation frequency and demand stability coefficient after removing outliers as the preset operation frequency and preset demand stability coefficient, respectively. In this invention, the preset operation frequency is 30 times / month and the preset demand stability coefficient is 0.6.

[0078] Specifically, the portrait optimization unit responds to the first optimization processing condition and determines the portrait supplementation method based on the comparison result between the effective demand difference and the preset effective demand difference;

[0079] If the effective demand difference is greater than the preset effective demand difference, the profile will be supplemented based on related product categories.

[0080] If the effective demand difference is less than or equal to the preset effective demand difference, the profile will be supplemented based on the relevant users.

[0081] The first optimization condition is to determine the portrait optimization method as being based on the difference in effective demand.

[0082] Effective demand difference = Demand stability coefficient - Preset demand stability coefficient;

[0083] The preset effective demand difference value can be adaptively set by the user according to the actual application scenario. It can be understood that the comparison between the effective demand difference and the preset effective demand difference determines the profile supplementation method as either related category supplementation or related user supplementation. Related category supplementation refers to products whose keyword repetition is greater than the preset repetition as related categories. Related products are arranged in descending order of keyword repetition to obtain a related sequence. The first three related products in the related sequence are then added to the related category and added to the user profile tag set. When users search for products on the online shopping platform, the platform will first recommend the target product and related products, expanding the user's selection range. The higher the user's requirement for profile supplementation accuracy, the smaller the preset effective demand difference value. This invention provides a method for setting the preset effective demand difference value by extracting the corresponding effective demand difference values ​​from the historical records that meet user needs, filtering out outliers, and recording the average of the effective demand differences after removing outliers as the preset effective demand difference value.

[0084] The TextRank algorithm is used to construct a word network graph from the text, and word importance scores are calculated through iterative voting. Candidate words are then sorted in descending order of their scores, and the top few candidate words are recorded as keywords. Keyword repetition = number of keywords that are the same as the target product keywords / total number of target product keywords.

[0085] The confirmation method for supplementing relevant users is as follows: The correlation coefficient between other users and the target user is detected, and the user with the highest correlation coefficient is recorded as the relevant user. The user profile tag set corresponding to the relevant user is extracted, and this set is added to the target user's user profile tag set. Here, the correlation coefficient = click-through rate + purchase frequency. The click-through rate and purchase frequency of other users for the target product are detected. Click-through rate = number of clicks / total number of clicks, and purchase frequency is the number of purchases within the monitoring period / monitoring period duration. In the calculation of the correlation coefficient, click-through rate and purchase frequency are calculated dimensionlessly, and the target product is the product purchased by the target user.

[0086] Specifically, the image optimization unit responds to the second optimization processing condition, determines the reference data state based on the image and text stability coefficient and the bias stability coefficient corresponding to the reference source, and determines the optimization assistance method based on the reference data state.

[0087] If the reference data state shows that the image-text stability coefficient is greater than the preset image-text stability coefficient and the bias stability coefficient is greater than the preset bias stability coefficient, then the optimization assistance method is image optimization.

[0088] If the reference data shows that the stability coefficient of the image and text is greater than the preset stability coefficient of the image and text and the emphasis stability coefficient is less than or equal to the preset emphasis stability coefficient, then the optimization assistance method is keyword optimization.

[0089] The second optimization condition is to determine the portrait optimization method as being based on the status of the reference data.

[0090] Specifically, the reference source corresponding to the most recent monitoring period of the target user is extracted. The reference source is the purchase interface corresponding to the product purchased in the most recent monitoring period. The purchase interface of the product corresponding to the target platform is recorded as the display interface. The display interface includes, but is not limited to, the product's image, price, and text description.

[0091] Image-text stability coefficient = α1 × image proportion + α2 × text proportion; where α1 is the first weight coefficient and α2 is the second weight coefficient. The values ​​of α1 and α2 can be set directly by the user based on domain experience, or by using statistical methods such as regression analysis or principal component analysis (PCA) to determine the contribution of image proportion and text proportion to the image-text stability coefficient, thereby determining the corresponding weight coefficient values. The greater the contribution, the greater the weight coefficient value. The weights can also be adjusted through training based on historical records (such as machine learning).

[0092] For a single display interface, its image proportion = product area / display interface area; text proportion = text area / display interface area; where the text area is the area of ​​the smallest rectangle that can completely cover the text description corresponding to each display interface. The product area is the area of ​​the target product in the product image. The product area is segmented using edge detection technology, and the product area is calculated using the contourArea function in OpenCV. This is easily understood by those skilled in the art and will not be elaborated here. The display interface area = length of the display interface × width of the display interface.

[0093] The image proportion and text proportion of the reference source correspond to the average image proportion and text proportion of the display interface, respectively. The bias stability coefficient is 1 / |image proportion - text proportion|.

[0094] Users can adaptively set the preset image and text stability coefficient and the preset bias stability coefficient according to actual application needs. It is understood that the greater the influence of the reference data status on the optimization assistance method, the smaller the preset image and text stability coefficient and the preset bias stability coefficient will be. This invention provides a method for setting the preset image and text stability coefficient and the preset bias stability coefficient. The image and text stability coefficient and the bias stability coefficient corresponding to the historical records that meet the user's needs are extracted, outliers are filtered out, and the average values ​​of the image and text stability coefficient and the bias stability coefficient after removing outliers are recorded as the preset image and text stability coefficient and the preset bias stability coefficient, respectively.

[0095] Specifically, the image analysis unit responds to the first optimization assistance mode condition, determines the degree of difference between the target image and the reference source based on the image diversity of the reference source and the proportion of the target image, and determines the portrait supplementation parameters based on the degree of difference between the target image.

[0096] If the difference in the target image is less than the preset difference in the target image, then the effective mean of the reference source is extracted and recorded as the image supplementation parameter for image supplementation.

[0097] If the difference in the target image is greater than or equal to the preset difference in the target image, the image supplementation parameters are determined based on the image matching coefficient.

[0098] The first optimization assistance method condition is that the optimization assistance method is confirmed to be image optimization.

[0099] The target image difference level = reference source image diversity / preset image diversity + target image proportion; where, the target image proportion = product area / reference source area; the average value of the target image difference level corresponding to the reference source is recorded as the effective mean. The effective mean is used as a profile supplementary parameter to supplement the user profile tag set. When recommending products in the future, products corresponding to the tags supplemented by the profile supplementary parameters will be given priority.

[0100] The method for determining image diversity is as follows: a two-dimensional coordinate system is established for the product area, where the x-coordinate and y-coordinate of any pixel block in the product area are both greater than or equal to 0, and the image diversity is denoted as L. ,in, This represents the pixel grayscale value at position (x, y) in the image. Q×W is the average of the pixel grayscale values ​​across all locations, where Q is the image resolution, Q is the horizontal coordinate length of the product area, and W is the vertical coordinate length of the product area.

[0101] The preset image diversity value can be adaptively set by the user according to the actual application requirements. It can be understood that the greater the influence of the image diversity of the reference source on the degree of difference of the target image, the smaller the preset image diversity value. This invention provides a method for setting the preset image diversity value, which extracts the corresponding image diversity in the historical records that meet the user's needs, filters out the outliers, and records the average value of the image diversity after removing the outliers as the preset image diversity.

[0102] The user can adaptively set the preset target image difference level according to the actual application scenario. It can be understood that the greater the impact of the target image difference level on the portrait supplementation parameters, the greater the preset target image difference level. This invention provides a method for setting the preset target image difference level by extracting the target image difference level corresponding to the historical records that meet the user's needs, filtering out outliers, and recording the average value of the target image difference level after removing outliers as the preset target image difference level.

[0103] Specifically, the image analysis unit responds to the preset supplementary parameter conditions and determines whether to supplement parameters based on the comparison result between the image matching coefficient and the preset image matching coefficient.

[0104] If the image matching coefficient is greater than the preset image matching coefficient, the average image matching coefficient will be used as a supplementary parameter for image supplementation.

[0105] If the image matching coefficient is less than or equal to the preset image matching coefficient, the image supplementation will fail.

[0106] The preset supplementary parameter condition is to determine the portrait supplementary parameters based on the image matching coefficient.

[0107] For a single reference source, its image matching coefficient = number of repeated keywords / total number of keywords; wherein, the images of each reference source are converted into text, and the text is recorded as image-converted text. The text conversion method is to standardize the image data through OpenCV image processing and background segmentation, and to detect the target image through YOLOv. ResNet is used to classify the target image by color and material and generate corresponding text descriptions, and these text descriptions are recorded as image-converted text. YOLOv and ResNet are content that is easy for those skilled in the art to understand, and will not be elaborated here.

[0108] Keywords that are the same as the keywords in the source text of the image are recorded as duplicate keywords; total keywords = keywords in the image-to-text + keywords in the source text.

[0109] The mean image matching coefficient is the average of the image matching coefficients of reference sources whose image matching coefficients are greater than the preset image matching coefficients. This mean image matching coefficient is used as a supplementary parameter to the user profile tag set. In subsequent product recommendations, products corresponding to tags supplemented by this parameter are prioritized.

[0110] The user can adaptively set the preset image matching coefficient value according to the actual application requirements. It is understood that the higher the user's requirement for the accuracy of the parameter supplement, the larger the preset image matching coefficient value. This invention provides a method for setting the preset image matching coefficient value by extracting the corresponding image matching coefficients in the historical records that meet the user's needs, filtering out the outliers, and recording the average value of the image matching coefficients after removing the outliers as the preset image matching coefficient.

[0111] Specifically, the text analysis unit responds to the conditions of the second optimization assistance method, determines the area category based on the dwell time and browsing speed, records the area with a browsing speed less than the preset browsing speed and a dwell time greater than the preset dwell time as the quick lookup dwell area, and records the text in the quick lookup dwell area as effective keywords;

[0112] The second optimization assistance method is determined by confirming that the optimization assistance method is keyword optimization.

[0113] For each reference source, the dwell time is the continuous display duration of the reference source, and the page scrolling speed is recorded as the browsing speed.

[0114] Users can adaptively set the preset browsing speed and preset dwell time values ​​according to their actual application needs. It is understood that the higher the user's requirement for the accuracy of the quick search dwell area, the larger the preset browsing speed and preset dwell time values ​​will be. This invention provides a method for setting preset browsing speed and preset dwell time values ​​by extracting the corresponding browsing speed and dwell time from the historical records that meet the user's needs, filtering out outliers, and recording the average values ​​of the browsing speed and dwell time after removing outliers as the preset browsing speed and preset dwell time, respectively.

[0115] Specifically, the text analysis unit determines the text structure coefficient based on the keyword diversity and keyword distribution density of effective keywords, and determines the profile supplementary parameters based on the difference in the text structure coefficients of the reference source.

[0116] If the difference in the text structure coefficient is less than or equal to the preset difference, then the supplementary parameter for the portrait is the text structure coefficient;

[0117] If the difference in the text structure coefficients is greater than the preset difference, the portrait supplementation will fail.

[0118] The text structure coefficient is calculated as follows: β1 × Keyword diversity + β2 × Keyword distribution density; where β1 is the first weighting coefficient and β2 is the second weighting coefficient. Keyword diversity is determined by extracting several keyword groups within the most recent monitoring period. Each keyword group consists of two randomly selected keywords. The average probability that the two keywords in the corresponding keyword groups are different is recorded as the keyword diversity. The number of keyword group extractions can be adaptively set by the user according to actual needs. It is understood that the higher the accuracy requirement for keyword diversity, the more keyword groups will be extracted. This invention provides a value for the number of keyword group extractions, which is 50 times. The keyword distribution density is determined by detecting the characters between two adjacent valid keywords, recorded as interval characters. Keyword distribution density = total number of interval characters / total number of text characters. The difference is calculated as: text structure coefficient - preset text structure coefficient.

[0119] Text structure coefficients are added to the user profile tag set as supplementary parameters for the profile. When recommending products in the future, products corresponding to the tags supplemented by the profile supplementary parameters will be recommended first.

[0120] The preset text structure coefficient can be set by the user according to the actual application needs. It can be understood that the higher the user’s requirements for the profile supplementation effect, the larger the preset text structure coefficient will be. This invention provides a way to set the preset text structure coefficient by extracting the corresponding text structure coefficient in the historical records that meet the user’s needs, filtering out the outliers, and recording the average value of the text structure coefficient after removing the outliers as the preset text structure coefficient.

[0121] The preset difference value can be set adaptively by the user according to the actual application needs. It can be understood that the higher the user's requirements for the effect of profile supplementation, the smaller the preset difference value will be. This invention provides a way to set the preset difference value by extracting the corresponding difference value in the historical records that meet the user's needs, filtering out the outliers, and recording the average difference value after removing the outliers as the preset difference value.

[0122] Please see Figure 4 The diagram illustrates a method for real-time acquisition and fusion of multi-source data according to the present invention. The present invention also provides a method for applying the aforementioned real-time acquisition and fusion system for multi-source data, comprising:

[0123] Periodically determine the accuracy status of the profiles for each target user based on profile coverage and the proportion of inefficient tags, and decide whether to optimize the profiles based on the accuracy status.

[0124] The corresponding profile optimization method is determined based on the target user's operation frequency and demand stability coefficient: optimization based on the effective demand difference or optimization based on the status of reference data.

[0125] When optimizing based on the effective demand difference, the method of supplementing the profile is determined based on the comparison result between the effective demand difference and the preset effective demand difference.

[0126] When performing optimization based on the status of reference data, the optimization assistance method is determined to be either image optimization or keyword optimization based on the image and text stability coefficient and the bias stability coefficient corresponding to the reference source.

[0127] During image optimization, the degree of difference between the reference source and the target image is determined based on the image diversity of the reference source and the proportion of the target image. The portrait supplementation parameters are then determined to be either the effective mean or the mean of the image matching coefficients based on the degree of difference of the target image.

[0128] When optimizing keywords, the text structure coefficient is determined based on the keyword diversity and keyword distribution density of effective keywords, and the difference between the text structure coefficients of the reference source and the text structure coefficients is used to determine whether the supplementary parameters of the profile are the difference of the text structure coefficients.

[0129] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A real-time acquisition and fusion system for multi-source data, characterized in that, include: The portrait analysis unit is used to determine the accuracy status of the portrait based on the portrait coverage rate and the proportion of inefficient tags. When the portrait accuracy status is less than or equal to the preset portrait coverage rate and the proportion of inefficient tags is greater than the preset inefficient tag proportion, it is determined that portrait optimization should be performed. The profile optimization unit, connected to the profile analysis unit, is used to determine the target user's operation status based on the target user's operation frequency and demand stability coefficient, and to determine the corresponding image optimization method based on the target user's operation status, either by optimizing based on the effective demand difference or by optimizing based on the reference data status; wherein, the profile supplementation method determined based on the effective demand difference is related category supplementation or related user supplementation; and the optimization assistance method determined based on the reference data status is image optimization or keyword optimization. An image analysis unit, which is connected to the image analysis unit and the image optimization unit respectively, is used to determine the degree of difference between the target image corresponding to the reference source based on the image diversity of the reference source and the proportion of the target image, and to determine the image supplementation parameters as the effective mean or the mean of the image matching coefficient based on the degree of difference of the target image. The text analysis unit, which is connected to the portrait analysis unit, the portrait optimization unit and the image analysis unit respectively, is used to determine the text structure coefficient based on the diversity and distribution density of effective keywords, and to determine the portrait supplementary parameters based on the difference in the text structure coefficients corresponding to the reference source. The profile optimization unit responds to preset optimization conditions, determines the user's operation status based on the target user's operation frequency and demand stability coefficient, and determines the corresponding profile optimization method based on the user's operation status. If the user's operation status is that the operation frequency is less than the preset operation frequency and the demand stability coefficient is greater than or equal to the preset demand stability coefficient, then the profile optimization method is to optimize based on the effective demand difference. If the user's operation status is that the operation frequency is greater than or equal to the preset operation frequency or the demand stability coefficient is less than the preset demand stability coefficient, then the profile optimization method is to optimize based on the reference data status. The preset optimization condition is determined by the portrait analysis unit to perform portrait optimization; The portrait optimization unit responds to the first optimization processing condition and determines the portrait supplementation method based on the comparison result between the effective demand difference and the preset effective demand difference; If the effective demand difference is greater than the preset effective demand difference, the profile will be supplemented based on related product categories. If the effective demand difference is less than or equal to the preset effective demand difference, the profile will be supplemented based on the relevant users. The first optimization condition is that the portrait optimization method is determined to be based on the difference in effective demand. The image optimization unit responds to the second optimization processing condition, determines the reference data status based on the image and text stability coefficient and the bias stability coefficient corresponding to the reference source, and determines the optimization assistance method based on the reference data status. If the reference data state shows that the image-text stability coefficient is greater than the preset image-text stability coefficient and the bias stability coefficient is greater than the preset bias stability coefficient, then the optimization assistance method is image optimization. If the reference data shows that the stability coefficient of the image and text is greater than the preset stability coefficient of the image and text and the emphasis stability coefficient is less than or equal to the preset emphasis stability coefficient, then the optimization assistance method is keyword optimization. The second optimization condition is that the portrait optimization method is determined to be based on the status of the reference data. If the difference in the target image is less than the preset difference in the target image, then the effective mean of the reference source is extracted and recorded as the image supplementation parameter for image supplementation. If the difference in the target image is greater than or equal to the preset difference in the target image, the image supplementation parameters are determined based on the image matching coefficient. If the difference in the text structure coefficient is less than or equal to the preset difference, then the supplementary parameter for the portrait is the text structure coefficient; If the difference in the text structure coefficient is greater than the preset difference, the portrait supplementation will fail. For a single monitoring cycle, the corresponding demand is stable. ;in, H represents the number of times the target user purchases the product tag i corresponding to the product, H represents the total number of purchases by the target user, and n represents the total number of product tags. Effective demand difference = Demand stability coefficient - Preset demand stability coefficient; Image-text stability coefficient = α1 × image proportion + α2 × text proportion; The biased stability coefficient = 1 / |image proportion - text proportion|; the average value of the difference between the target image and the reference source is recorded as the effective mean; The mean image matching coefficient is the average of the image matching coefficients of the reference sources whose image matching coefficients are greater than the preset image matching coefficients; The reference source is the purchase interface corresponding to the product purchased in the most recent monitoring period; Text structure coefficient = β1 × keyword diversity + β2 × keyword distribution density.

2. The real-time acquisition and fusion system for multi-source data according to claim 1, characterized in that, The profile analysis unit periodically determines the profile accuracy status of each target user based on profile coverage and the proportion of inefficient tags. When the profile accuracy status is that the profile coverage is less than or equal to the preset profile coverage and the proportion of inefficient tags is greater than the preset proportion of inefficient tags, it determines that profile optimization should be performed.

3. The real-time acquisition and fusion system for multi-source data according to claim 1, characterized in that, The image analysis unit responds to the first optimization assistance mode condition, determines the degree of difference between the target image and the reference source based on the image diversity and the proportion of the target image, and determines the image supplementation parameters based on the degree of difference between the target image. The first optimization assistance method condition is that the optimization assistance method is confirmed to be image optimization.

4. The real-time acquisition and fusion system for multi-source data according to claim 1, characterized in that, The image analysis unit responds to the preset supplementary parameter conditions and determines whether to supplement parameters based on the comparison result between the image matching coefficient and the preset image matching coefficient. If the image matching coefficient is greater than the preset image matching coefficient, the average image matching coefficient will be used as a supplementary parameter for image supplementation. If the image matching coefficient is less than or equal to the preset image matching coefficient, the image supplementation will fail. The preset supplementary parameter condition is to determine the portrait supplementary parameters based on the image matching coefficient.

5. The real-time acquisition and fusion system for multi-source data according to claim 1, characterized in that, The text analysis unit responds to the second optimization assistance condition, determines the area category based on the dwell time and browsing speed, and records the area corresponding to the browsing speed being less than the preset browsing speed and the dwell time being greater than the preset dwell time as the quick lookup dwell area, and records the text in the quick lookup dwell area as effective keywords; The second optimization assistance method is determined by confirming that the optimization assistance method is keyword optimization.

6. The real-time acquisition and fusion system for multi-source data according to claim 1, characterized in that, The text analysis unit determines the text structure coefficient based on the keyword diversity and keyword distribution density of effective keywords, and determines the profile supplementary parameters based on the difference in the text structure coefficients of the reference source.

7. A method for real-time acquisition and fusion of multi-source data as described in any one of claims 1 to 6, characterized in that, include: Periodically determine the accuracy status of the profiles for each target user based on profile coverage and the proportion of inefficient tags, and decide whether to optimize the profiles based on the accuracy status. The corresponding profile optimization method is determined based on the target user's operation frequency and demand stability coefficient: optimization based on the effective demand difference or optimization based on the status of reference data. When optimizing based on the effective demand difference, the method of supplementing the profile is determined based on the comparison result between the effective demand difference and the preset effective demand difference. When performing optimization based on the status of reference data, the optimization assistance method is determined to be either image optimization or keyword optimization based on the image and text stability coefficient and the bias stability coefficient corresponding to the reference source. During image optimization, the degree of difference between the reference source and the target image is determined based on the image diversity of the reference source and the proportion of the target image. The portrait supplementation parameters are then determined to be either the effective mean or the mean of the image matching coefficients based on the degree of difference of the target image. When optimizing keywords, the text structure coefficient is determined based on the keyword diversity and keyword distribution density of effective keywords, and the difference between the text structure coefficients of the reference source and the text structure coefficients is used to determine whether the supplementary parameters of the profile are the difference of the text structure coefficients.

Images