Internet shopping mall operation management system based on big data

Abstract

The invention provides an internet shopping mall operation management system based on big data, and relates to the field of operation management, and the system comprises a deep buyer updating module, a recommendation acceptability determination module, a friend recommendation degree determination module, a friend recommendation degree updating module and a recommendation state matching module. The deep buyer updating module is used for creating a deep buyer library; a recommendation acceptability determination module determines the recommendation acceptability of the buyer for each target friend according to the historical chat information and the historical purchase information of the buyer; the friend recommendation degree determination module determines the friend recommendation degree; the friend recommendation degree updating module updates the friend recommendation degree mending the commodity recently based on the search text; and the recommendation state matching module judges the recommendation state of the recently mentioned commodity according to the friend recommendation degree. According to the system, commodity mention information in social interaction can be effectively mined and utilized, and the matching degree between recommended content and real demands of a user is effectively improved; personalized requirements of users are met, and excessive pushing is avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of operation management, in particular to an Internet mall operation management system based on big data. BACKGROUND

[0002] With the development of the Internet, Internet malls have also emerged, and the development status of Internet malls presents the characteristics of diversification, intelligence and globalization. At present, e-commerce platforms have penetrated into various fields of daily life, covering a wide range of categories from physical goods to digital services. Mainstream platforms continuously improve user experience and conversion efficiency through algorithm recommendation, live streaming, social e-commerce and other innovative modes.

[0003] In recent years, the personalized recommendation system in the Internet mall has become a key means to improve user shopping experience and promote consumption conversion. Traditional recommendation systems mainly rely on user historical browsing and purchase behavior, combined with collaborative filtering, content recommendation or deep learning algorithms to generate a recommendation list. However, the existing technology still has the following main defects: traditional recommendation systems mainly rely on user's own shopping behavior data, ignoring the influence of social relationships on shopping decisions. For example, users may have a purchase intention due to a friend's recommendation, but existing systems cannot effectively mine and utilize product mention information in social interactions, resulting in low matching degree of recommended content to user's real needs. Existing recommendation systems usually default to obtain user behavior data, but lack a dynamic permission management mechanism. For example, some users may be willing to accept deep recommendations, but the system cannot dynamically adjust the recommendation strategy according to their long-term interaction value (such as retention score), resulting in loss of high-value users or excessive disturbance of low-value users. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the present application is realized by the following technical scheme: An Internet mall operation management system based on big data, the system comprising: a deep buyer updating module for receiving a deep push request uploaded by a buyer, creating a deep buyer library, periodically sending a permission acquisition request to the buyers in the deep buyer library, and periodically updating the buyers in the deep buyer library based on a preset dynamic evaluation model; a recommendation acceptance determination module for obtaining historical chat information of the buyers in the deep buyer library and target friends, and determining the recommendation acceptance of the buyers for each target friend according to the historical chat information and the historical purchase information of the buyers; The friend recommendation degree determination module is used to obtain recent chat information between buyers and target friends in the deep buyer database within the monitoring period, extract recently mentioned product information based on the recent chat information, including the person who actively mentioned it, product type and product model, and count the number of times each target friend mentioned the recently mentioned product. Based on the number of mentions and the buyer's acceptance of the recommendation of each target friend, the friend recommendation degree of the recently mentioned product is determined. The friend recommendation update module is used to receive the search text entered by the buyer in the search box of the mall in real time, and update the friend recommendation score of recently mentioned products based on the search text; The recommendation status matching module is used to determine the recommendation status of recently mentioned products based on the friend recommendation level. The recommendation status includes blocked status, push ready status, and push immediately status.

[0005] Preferably, the deep buyer update module includes: The Deep Buyer Database Creation Unit is used to receive deep push requests uploaded by buyers and create a deep buyer database. The permission request acquisition unit is used to periodically send permission acquisition requests to buyers in the deep buyer database. The behavior log acquisition unit is used to acquire the buyer's behavior log within a unit period. The behavior log includes the number of times the buyer clicks on recommended products, the number of times the buyer purchases recommended products, the number of times the buyer marks products as uninteresting, and the number of times the buyer manually closes recommendations. The exposure count acquisition unit is used to acquire the number of times a recommended product is exposed within a unit period. The retention score calculation unit is used to input the behavior logs and recommended exposure counts into a preset dynamic evaluation model and output the buyer's retention score. The retention score determination unit is used to determine whether to remove the buyer from the deep buyer database based on the retention score.

[0006] Preferably, in the retention scoring determination unit, the content for determining whether to remove a buyer from the deep buyer database based on the retention score includes: When the retention score is greater than the first preset score threshold, it is determined that the buyer is retained in the deep buyer database and deep recommendations continue to be provided. When the retention score is not greater than the first preset score threshold but greater than the second preset score threshold, it is determined that the buyer is retained in the deep buyer database, and a reminder message is sent to the buyer at the same time. When the retention score is not greater than the second preset score threshold, it is determined that the buyer will be removed, that is, the buyer will be removed from the deep buyer database.

[0007] Preferably, the recommendation acceptance determination module includes: The historical chat information acquisition unit is used to acquire historical chat information between buyers and target friends in the deep buyer database; A historical mention information acquisition unit is used to extract historically mentioned product information from the historical chat information; The historical purchase information acquisition unit is used to acquire information about the buyer's historical purchases. The recommendation acceptance determination unit is used to determine the buyer's recommendation acceptance for each target friend based on the historical mentioned product information and historical purchased product information.

[0008] Preferably, in the recommendation acceptance determination unit, the content for determining the buyer's recommendation acceptance for each target friend based on the historical mentioned product information and historical purchased product information includes: The product type and model in the historically mentioned product information are compared with the product type and model in the historically purchased product information. If the product type and model are the same, the matching degree is determined to be 1. x 1; If only the product types are the same, the matching degree is determined to be... x 2; If both the product type and product model are different, the matching degree is determined to be... x 3; If the buyer's past purchase history includes items mentioned by the target friend, the sum of the matching scores is used to obtain the additional matching score. The final matching degree is obtained based on the additional matching degree and the preset basic matching degree; Convert the final match score of all target friends into a recommendation acceptance score of 0 to 1.

[0009] Preferably, the friend recommendation degree determination module includes: The recent information acquisition unit is used to acquire recent chat information between buyers and target friends in the deep buyer database within the monitoring period; The product information extraction unit is used to extract recently mentioned product information based on the recent chat information; The mention count unit is used to count the number of times each target friend mentions a recently mentioned product; The recommendation calculation unit is used to determine the friend recommendation score of each recently mentioned product based on the number of mentions and the buyer's acceptance of recommendations from each target friend.

[0010] Preferably, in the mention count unit, when counting the number of times each target friend mentions a recently mentioned product, semantic analysis is performed on the chat information containing the recently mentioned product to determine whether it is a valid mention. If it is a valid mention, the chat information is checked to see if it is mentioned repeatedly. If not, the mention count is incremented by 1.

[0011] Preferably, the friend recommendation update module includes: The information extraction unit is used to receive the search text entered by the buyer in the search box of the mall in real time, and extract the search product information in the search text, including product type and product model information; The model information determination unit is used to determine whether product model information exists in the searched product information. The first recommendation update unit is used to determine that the friend recommendation level remains unchanged when there is no product model information in the search product information. The similar product extraction unit is used to extract recently mentioned products of the same type based on the product type of the searched product when product model information exists in the searched product information; The second recommendation update unit is used to match the product model of the searched product with the product model of the recently mentioned products of the same type when there are recently mentioned products of the same type as the searched product, and to determine whether there is a target recently mentioned product that successfully matches the product model of the searched product. The target recently mentioned product refers to a recently mentioned product that successfully matches the product model of the searched product. When there is a target recently mentioned product that successfully matches the product model of the searched product, the friend recommendation score of the target recently mentioned product is updated. The third recommendation update unit is used to update the friend recommendation scores of all recently mentioned products of the same type as the searched product when there are no target recently mentioned products that successfully match the model of the searched product. The fourth update unit for recommendation score is used to determine that the friend recommendation score remains unchanged when there are no recently mentioned products of the same type as the searched product.

[0012] Preferably, in the second update unit of the recommendation degree, the definition of successful matching is: calculating the similarity between product models and comparing it with a preset similarity threshold. When the calculated similarity exceeds the preset similarity threshold, it is determined to be a successful match.

[0013] Preferably, the recommended status matching module includes: The immediate push confirmation unit is used to determine the recommendation status of recently mentioned products as immediate push status when the friend recommendation degree is greater than the first preset recommendation threshold. The recommendation preparation determination unit is used to determine the recommendation status of recently mentioned products as the recommendation preparation status when the friend recommendation degree is greater than the second preset recommendation threshold and not greater than the first preset threshold. The push notification blocking determination unit is used to determine that the recommendation status of recently mentioned products is blocked when the friend's recommendation degree is not greater than a preset second preset recommendation threshold.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention provides an internet e-commerce operation and management system based on big data. This system effectively mines and utilizes product mention information from social interactions, significantly improving the matching degree between recommended content and users' actual needs. Firstly, this invention distinguishes e-commerce users into ordinary users and deeply pushed users, thus meeting users' personalized needs while avoiding excessive push notifications. Secondly, this invention quantifies social trust by objectively measuring the credibility / acceptance of friend recommendations through the matching degree between historical product mentions and purchase records. Combining mention frequency (behavioral frequency) and recommendation acceptance (quality weight) avoids misjudgments based solely on chat frequency and reduces user aversion. This invention also strengthens friend recommendations based on search, improving push accuracy. Attached Figure Description

[0015] Figure 1 This is a structural block diagram of the Internet e-commerce operation and management system based on big data provided in the embodiments of the present invention. Detailed Implementation

[0016] To make the technical problems to be solved, the technical solutions, and the beneficial effects of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention.

[0017] Please see Figure 1 This invention provides an internet e-commerce operation and management system based on big data. The system 10 includes: The deep buyer update module 11 is used to receive deep push requests uploaded by buyers / users, create a deep buyer database, periodically send permission acquisition requests to buyers in the deep buyer database, and periodically update the buyers in the deep buyer database based on a preset dynamic evaluation model; periodically sending permission acquisition requests to buyers in the deep buyer database means requesting buyers to grant information access permissions; the deep buyer database includes buyer ID information, buyer friend ID information, and chat information between the buyer and the target friend; in this invention, mall users are first divided into ordinary users and deep push users, thereby achieving the goal of meeting users' personalized needs while avoiding excessive push. In this embodiment, the deep buyer update module 11 includes: The Deep Buyer Database Creation Unit is used to receive deep push requests uploaded by buyers and create a deep buyer database. The permission request acquisition unit is used to periodically send permission acquisition requests to buyers in the deep buyer database; by periodically sending permission acquisition requests, it is clear that users agree to the deep recommendation service, which not only protects user privacy and compliance, but also filters out users who are truly willing to accept deep recommendations. The behavior log acquisition unit is used to acquire the buyer's behavior log within a unit period. The behavior log includes the number of times the buyer clicks on recommended products, the number of times the buyer purchases recommended products, the number of times the buyer marks products as uninteresting, and the number of times the buyer manually closes recommendations. The unit period is, for example, six months or one year. Based on the dynamic evaluation model of behavior logs (such as clicks, purchases, marking products as uninteresting, etc.) and recommendation exposure times, a retention score is output to objectively quantify the buyer's dependence on and activity level of the recommendation service. The exposure count acquisition unit is used to obtain the number of times a recommended product is displayed within a unit period. The number of recommended exposures is the number of times a recommended product is displayed. Each time a recommended product is displayed on the buyer's interface, an exposure log is triggered. Each exposure log usually includes user ID, product information, exposure time, etc. The dwell time is ≥500ms (to avoid invalid exposures caused by rapid scrolling). If the same buyer sees the same product multiple times in the same session / display interface, it is only counted as 1 exposure (to prevent duplicate counting). The retention score calculation unit is used to input the behavior logs and recommendation exposure counts into a preset dynamic evaluation model and output the buyer's retention score. The dynamic evaluation model can calculate the retention score based on click-through rate (CTR), purchase conversion rate (PCR), and negative feedback intensity. Specifically, CTR = number of times the buyer clicks on recommended products / number of recommendation exposures; PCR = number of times the buyer purchases recommended products / number of clicks; and negative feedback intensity = number of times the buyer marks the product as uninterested + number of times the buyer manually closes the recommendation. Each of CTR, PCR, and negative feedback intensity has a corresponding weight coefficient, and typically the weight coefficient for CTR > PCR > 0 > negative feedback intensity. The dynamic evaluation model calculates the retention score based on a weighted summation. The retention score determination unit is used to determine whether to remove a buyer from the deep buyer database based on the retention score; it automatically removes low retention buyers through retention score determination rules (such as threshold comparison) to avoid wasting resources; it sends reminders to low and medium retention buyers to encourage them to re-engage; and it retains high retention buyers and continues to provide in-depth recommendations to form a virtuous cycle of user groups.

[0018] The retention scoring determination unit includes determining whether to remove a buyer from the deep buyer database based on the retention score, including: When the retention score is greater than the first preset score threshold, it is determined that the buyer is retained in the deep buyer database and will continue to receive in-depth recommendations. The first preset score threshold > the second preset score threshold > 0. The first preset score threshold (high threshold) and the second preset score threshold (low threshold) are not set arbitrarily. They can be determined by first calculating the percentile of the retention scores of all users and then using the percentiles. For example, the first preset score threshold can be set to the lowest score of the top 20% of users, and the second preset score threshold can be set to the highest score of the bottom 30% of users. When the retention score is not greater than the first preset score threshold but greater than the second preset score threshold, it is determined that the buyer is retained in the deep buyer database, and a reminder message is sent to the buyer. The reminder message includes inquiries about the buyer's user experience and whether to adjust preferences. When the retention score is not greater than the second preset score threshold, the buyer is removed from the deep buyer database. A three-level retention scoring mechanism (retention / reminder / removal) is used to achieve intelligent updates of the user database.

[0019] The deep buyer update module 11 in this invention achieves precise user screening: it automatically identifies high-value deep buyers through a preset dynamic evaluation model (click-through rate, purchase conversion rate, negative feedback intensity), avoiding invalid recommendations that harass ordinary users. Simultaneously, this invention periodically requests authorization to ensure the legality of data collection and compliance with privacy regulations such as GDPR. It automatically removes inactive users (retention score ≤ second preset score threshold), reducing the load on invalid data processing.

[0020] The recommendation acceptance determination module 12 is used to obtain the historical chat information between the buyer and the target friends in the deep buyer database, and determine the buyer's recommendation acceptance for each target friend based on the historical chat information and the buyer's historical purchase information; the historical chat information refers to the chat information within the first historical interval, and the historical purchase information refers to the purchase information within the second historical interval. Usually, the length of the second historical interval is greater than or equal to the length of the first historical interval, and the starting points of the first and second historical intervals are the same. The ending point of the second historical interval is a preset time, which may be the current time or a time before today. In this embodiment, the recommendation acceptance determination module 12 includes: Historical chat information acquisition unit, which is used to acquire the historical chat information between the buyers in the deep buyer database and the target friends; where the target friends can be friends with a closeness exceeding a preset intimacy threshold, and the higher the closeness, the closer the relationship between the corresponding user and the friend. The closeness can be calculated by weighted summation of elements such as the chat frequency within a unit time, the proportion of common friends (number of common friends / total number of the user's friends), and gift exchange records, etc. This is the prior art and will not be elaborated in the embodiments of the present invention; Historical mention information acquisition unit, which is used to extract historical mentioned commodity information from the historical chat information. The historical mentioned commodity information includes the extracted commodity type, commodity model, mentioned friend ID, mention time, etc.; The main steps for extracting historical mentioned commodity information from the historical chat information include: 1) Text cleaning: removing irrelevant symbols (such as emoticons, URLs, etc.); unifying simplified and traditional Chinese (normalizing "iPhone" and "愛瘋"); 2) Sentence splitting and word segmentation: using NLP tools (such as jieba, NLTK) to split sentences and keywords; matching the keywords with a preset keyword library to determine the commodity type and commodity model; This is the prior art and will not be specifically elaborated in this embodiment; Historical purchase information acquisition unit, which is used to acquire the historical purchased commodity information of the buyer. The historical purchased commodity information includes the purchased commodity type, commodity model, purchase time, etc.; Recommendation acceptance determination unit, which is used to determine the recommendation acceptance of the buyer for each target friend based on the historical mentioned commodity information and the historical purchase commodity information.

[0021] In the recommendation acceptance determination unit, the content of determining the recommendation acceptance of the buyer for each target friend based on the historical mentioned commodity information and the historical purchase commodity information includes: Comparing the commodity type and commodity model in the historical mentioned commodity information with the commodity type and commodity model in the historical purchase commodity information. If both the commodity type and the commodity model are the same, the matching degree is determined to be x 1, x The value of 1 ranges from 0.8 to 1.0, usually 1.0; if only the commodity type is the same, the matching degree is determined to be x 2, x The value of 2 ranges from 0.4 to 0.6, usually 0.5; if both the commodity type and the commodity model are different, the matching degree is determined to be x 3, x The value of 3 is 0; A time decay factor can also be introduced to make the weight of recently mentioned commodities higher; Extract the matching degree (exact match, type match or no match) between the commodities mentioned by friends in the historical chat and the commodities actually purchased by the buyer, which intuitively reflects the influence of friend recommendations on the buyer's purchase decision; If the buyer's past purchase history includes items mentioned by the target friend, the sum of the matching scores is used to obtain the additional matching score. The final matching score is obtained based on the additional matching score and the preset basic matching score. The basic matching score is usually 1. For example, if three products recommended by friend A are purchased, the additional matching score is: 1.0 (the first recommended product completely matches a previously purchased product) + 0.5 (the second recommended product only matches a previously purchased product by type) + 1.0 (the third recommended product completely matches a previously purchased product). Friend A's final matching score = 1.0 + (1.0 + 0.5 + 1.0) = 3.5. Combining multiple effective recommendations (additional matching scores) with the preset basic matching score, the final recommendation acceptance score (range 0~1) is generated to avoid the randomness of a single interaction and improve the stability of acceptance score evaluation. The final match scores of all target friends are converted into recommendation acceptance scores ranging from 0 to 1. The user's recommendation acceptance score for friend A is calculated as: Friend A's final match score / the highest final match score among all target friends. For example, if friend A's final match score is 3.5 and the highest final match score among all target friends is 5.0, then the user's recommendation acceptance score for friend A is 0.7. An independent recommendation acceptance score is generated for each target friend, achieving a precise association between "friend-product-acceptance score" and providing a differentiated basis for subsequent recommendation strategies.

[0022] In this invention, the recommendation acceptance determination module 12 objectively measures the credibility of friend recommendations by comparing the matching degree between historically mentioned products and purchase records. The normalized output compresses the recommendation acceptance to a range of 0-1, facilitating cross-friend comparisons.

[0023] The friend recommendation score determination module 13 is used to obtain recent chat information between buyers and target friends in the deep buyer database within the monitoring period, extract recently mentioned product information based on the recent chat information, including the person who actively mentioned the product, the product type and the product model, and count the number of times each target friend mentions the recently mentioned product. The friend recommendation score is determined based on the number of mentions and the buyer's acceptance of the recommendation of each target friend. The friend recommendation score of the recently mentioned product is obtained by weighted summation based on the buyer's acceptance of the recommendation of each target friend and the number of mentions of each target friend. In this embodiment, the friend recommendation degree determination module 13 includes: The recent information acquisition unit is used to acquire recent chat information between buyers and target friends in the deep buyer database within the monitoring period; The product information extraction unit is used to extract recently mentioned product information based on the recent chat information. Recently mentioned product information refers to product information mentioned in the chat history between the buyer and the target friend within the monitoring period, including the person who actively mentioned it, the product type, and the product model. The mention count unit is used to count the number of times each target friend mentions a recently mentioned product; The recommendation score calculation unit is used to determine the friend recommendation score of each recently mentioned product based on the number of mentions and the buyer's acceptance of recommendations from each target friend. Specifically, it calculates the friend recommendation score of a recently mentioned product by weighting and summing the buyer's acceptance of recommendations from each target friend and the number of mentions from each target friend. For example, if a recently mentioned product has been mentioned and recommended by friends A, B, and C, then the friend recommendation score of this recently mentioned product = a*Y1 + b*Y2 + c*Y3, where a represents the number of times friend A mentioned the product, Y1 represents the seller's acceptance of friend A's recommendation, b represents the number of times friend B mentioned the product, Y2 represents the seller's acceptance of friend B's recommendation, c represents the number of times friend C mentioned the product, and Y3 represents the seller's acceptance of friend C's recommendation. The mentioned product information includes product type (e.g., mobile phone, computer, clothing, etc.) and product model (e.g., iPhone). 14) In this invention, standardized product information is stored in a product type-model storage library to establish a mapping relationship between product model and its corresponding product type. The product type-model storage library stores several pairs of product types and product models. The product type can be deduced from the product model, but the product model cannot be deduced from the product type. The recent chat information refers to the chat information within the monitoring period. The monitoring period refers to the time interval with the current time as the end point and a preset duration as the interval length, which can usually refer to the first half of the year. In this invention, the recommendation acceptance of friends and the number of times the product has been mentioned recently are combined to generate the recommendation degree of friends for specific products. This takes into account both the historical trust foundation and the current interaction popularity, thereby improving the timeliness and accuracy of the recommendation degree.

[0024] In the mention count unit, when counting the number of times each target friend mentions a recently mentioned product, semantic analysis is performed on chat messages containing recently mentioned products to determine whether they are valid mentions (e.g., "Don't buy Huawei Mate 60" is not counted). The semantic analysis of chat messages containing recently mentioned products to determine whether they are valid mentions mainly includes sentiment polarity judgment (negative sentence recognition) and intent recognition. Sentiment polarity judgment (negative sentence recognition) can be achieved using a sentiment analysis model or rule base, analyzing the sentiment tendency of sentences or contexts containing product names. If obvious negative words (e.g., "Don't buy," "Don't recommend," "Dislike," "Disappointing") and negative sentiment are identified, it is determined as an invalid mention; if positive words (e.g., "Want to buy," "Recommend," "I'm interested") and positive or neutral sentiment are identified, it is determined as a valid mention. Intent recognition can be achieved using an intent classification model to determine whether the user's statement is a "recommendation," "inquiry," "complaint," or "simple mention." Only positive or neutral intents such as "recommendation" and "inquiry" are counted as valid mentions. Negative intents such as "complaint" are filtered out. The step of performing semantic analysis on chat messages containing recently mentioned products to determine whether they are valid mentions is existing NLP technology and will not be elaborated here. If it is a valid mention, check whether it is mentioned repeatedly in the chat message (repeated mentions of the same product in the same session are only counted once to prevent count manipulation; a specific limitation method could be: multiple discussions of the same product by the same target friend within 30 minutes are considered as one mention). If not, the mention count is incremented by 1.

[0025] In this invention, the friend recommendation degree determination module 13 uses multi-dimensional weighting, which combines the number of mentions (behavioral frequency) and recommendation acceptance (quality weight) to avoid misjudgment based solely on chat frequency; by responding to buyer search behavior in real time, it dynamically adjusts the friend recommendation degree to ensure that the recommendation is synchronized with the user's current needs.

[0026] The friend recommendation update module 14 is used to receive search text entered by buyers in the mall's search box in real time, and update the friend recommendation score of recently mentioned products based on the search text; it determines the recommendation status (including immediate push status, recommendation preparation status, and blocked status) through thresholds to ensure that products with high friend recommendation scores are pushed first, products with medium friend recommendation scores are on standby, and products with low friend recommendation scores are blocked to avoid excessively disturbing users; it blocks pushes of products with low acceptance to reduce user resistance to irrelevant recommendations; the recommendation preparation status reserves a push window for products of potential interest, balancing recommendation coverage and user tolerance; it concentrates push resources on high-value products to improve recommendation conversion rates; and at the same time, through status layering, it provides operators with a visual recommendation priority to assist in decision optimization.

[0027] In this embodiment, the friend recommendation update module 14 includes: The information extraction unit is used to receive the search text entered by the buyer in the search box of the mall in real time, and extract the search product information in the search text, including product type and product model information; The model information determination unit is used to determine whether product model information exists in the searched product information. The first recommendation update unit is used to determine that the friend recommendation level remains unchanged when there is no product model information in the search product information. The similar product extraction unit is used to extract recently mentioned products of the same type based on the product type of the searched product when product model information exists in the searched product information; The second recommendation update unit is used to match the model of the searched product with the model of the recently mentioned products of the same type when there are recently mentioned products of the same type as the searched product, and determine whether there is a target recently mentioned product that successfully matches the model of the searched product. When there is a target recently mentioned product that successfully matches the model of the searched product, the friend recommendation score of the target recently mentioned product is updated. The update method is a weighted update. At this time, the friend recommendation score is updated with weight based on the first update coefficient. The third recommendation update unit is used to update the friend recommendation scores of all recently mentioned products of the same type as the searched product when there are no target recently mentioned products that successfully match the model of the searched product. At this time, the friend recommendation scores are updated by weighting based on the second update coefficient, where the first update coefficient is greater than the second update coefficient. The fourth update unit for recommendation score is used to determine that the friend recommendation score remains unchanged when there are no recently mentioned products of the same type as the searched product.

[0028] In the second update unit of the recommendation, the definition of successful matching is: calculate the similarity between product models and compare it with a preset similarity threshold. When the calculated similarity exceeds the preset similarity threshold, it is determined to be a successful match.

[0029] The friend recommendation update module 14 in this invention realizes search intent enhancement, that is, when a user actively searches, the recommendation of related products is instantly improved through model similarity matching (such as "iPhone15"≈"iPhone15 Pro").

[0030] The recommendation status matching module 15 is used to determine the recommendation status of recently mentioned products based on the friend recommendation level. The recommendation status includes a blocked status, a push ready status, and an immediate push status.

[0031] In this embodiment, the recommendation status matching module 15 includes: An immediate push confirmation unit is used to determine the recommendation status of recently mentioned products as "immediate push" when the friend recommendation score is greater than a first preset recommendation threshold. The first preset recommendation threshold > second preset recommendation threshold > 0. The first and second preset recommendation thresholds are determined through historical A / B testing, observing the click-through rate and purchase conversion rate of products in different "friend recommendation score" ranges. The first preset recommendation threshold is typically set at the lower limit of the range that ensures a high conversion rate. For example, if testing shows that products with a friend recommendation score higher than 0.8 have a significantly higher purchase conversion rate than the average, then the first threshold can be set to 0.8. The goal is to ensure that "immediate pushes" are for high-value, high-probability-of-transaction products. The second preset recommendation threshold is typically set at the upper limit of the range where the click-through rate is acceptable, but the purchase conversion rate is close to the average or even lower. For example, products with a friend recommendation score lower than 0.3 have a very low click-through rate and a high user close rate, then the second threshold can be set to 0.3. The goal is to filter out products that are almost ineffective even when recommended and only bother users. The recommendation preparation determination unit is used to determine the recommendation status of recently mentioned products as the recommendation preparation status when the friend recommendation degree is greater than the second preset recommendation threshold and not greater than the first preset threshold. The push notification blocking unit is used to determine that the recommendation status of recently mentioned products is blocked when the friend's recommendation score is not greater than a preset second preset recommendation threshold. Blocking means that the corresponding recently mentioned products will not be pushed to this buyer. Applicable scenarios: Products are only mentioned once by chance (low weight); Push preparation state (background preparation): The system determines that the product has some recommendation value, but has not yet reached the standard for immediate push, so it prepares the content first (such as generating recommendation copy and loading product details); When the product is in the push preparation state and the user has recently had related behavior (such as browsing similar products), it is not pushed immediately, but displayed when the user visits the relevant page again (such as the product list page); Applicable scenarios: Products are mentioned 3 times, but the user has not yet shown clear interest (such as not searching for related terms); Immediate push state (real-time recommendation): The system determines that the user has a strong interest in the product and immediately displays it in the recommendation position; Applicable scenarios: Products are mentioned frequently (such as 10 times), or the user actively searches for the model.

[0032] The hierarchical push strategy in the recommendation status matching module 15 of this invention includes: Immediate push (> first preset recommendation threshold): Frequent mentions or search matching products to seize the user's decision window. Ready state (second preset recommendation threshold ~ first preset recommendation threshold): Preloaded content to be displayed when the user browses relevant pages. Blocked state (≤ second preset recommendation threshold): Filtering occasional mentions to reduce interference. This invention reduces invalid recommendations, avoids user aversion, and effectively improves user satisfaction. The preset recommendation threshold in this embodiment can also be set to be scene-adaptive: the threshold for high-priced products (such as mobile phones) is automatically lowered, while the threshold for daily necessities is raised to balance the recommendation intensity.

[0033] This invention creatively integrates users' social chat history (including historical and recent chat messages) with e-commerce recommendation systems, breaking through the limitations of traditional recommendation systems that rely solely on users' own behavioral data. By analyzing product mentions between users and their friends, a new recommendation metric, "friend recommendation degree," is established. This allows the recommendation system to more accurately capture the impact of social relationships on shopping decisions, effectively improving the match between recommended content and users' actual needs.

[0034] The present invention also designed a multi-level state matching mechanism (blocking state, push preparation state, and immediate push state) to realize the dynamic adjustment of the recommendation strategy.

[0035] By monitoring users' input behavior in the search box in real time, a recommendation update mechanism with instant feedback was established, which significantly improved the timeliness and accuracy of recommendations.

[0036] The concept of a "deep buyer database" was proposed, which uses a dynamic evaluation model to intelligently filter and manage users. A retention scoring system with multiple dimensions, including behavioral log analysis and exposure statistics, was designed to achieve accurate evaluation and dynamic adjustment of user value.

[0037] In practical applications, when online shopping malls push products to users, they not only include the friend recommendation rate mentioned in this invention, but also other additional elements, such as the product's sales volume, positive review rate, and promotion fees. Then, a weighted calculation is performed based on multiple elements to determine the final push content (push order). However, this is not the focus of innovation in this invention, so it will not be described in detail in the embodiments of this invention.

[0038] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0039] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0040] Those skilled in the art will also understand that the various exemplary logic blocks, modules, circuits, and algorithm steps described in connection with the present application can be implemented as electronic hardware, computer software, or a combination of both.

[0041] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems and methods according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that indicated in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved.

[0042] It should also be noted that each block in a block diagram and / or flowchart, as well as combinations of blocks in a block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0043] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A big data-based Internet mall operation management system, characterized by, The system comprises: a deep buyer updating module configured to receive a deep push request uploaded by a buyer, create a deep buyer library, periodically send a permission acquisition request to the buyers in the deep buyer library, and periodically update the buyers in the deep buyer library based on a preset dynamic evaluation model; a recommendation acceptance determination module configured to acquire historical chat information of the buyers in the deep buyer library and target friends, and determine a recommendation acceptance of the buyers to the target friends according to the historical chat information and historical purchase information of the buyers; a friend recommendation degree determination module configured to acquire recent chat information of the buyers in the deep buyer library and the target friends in a monitoring interval, extract recent mentioned commodity information according to the recent chat information, and count a mentioning frequency of the recent mentioned commodity by the target friends, and determine a friend recommendation degree of the recent mentioned commodity according to the mentioning frequency and the recommendation acceptance of the buyers to the target friends; a friend recommendation degree updating module configured to receive a search text input in a search box of a mall by the buyer in real time, and update the friend recommendation degree of the recent mentioned commodity based on the search text; a recommendation state matching module configured to determine a recommendation state of the recent mentioned commodity according to the friend recommendation degree, wherein the recommendation state comprises a shielding state, a push preparation state and an immediate push state.

2. The big data-based Internet mall operation management system according to claim 1, characterized by, The deep buyer updating module comprises: a deep buyer library creating unit configured to receive a deep push request uploaded by a buyer, and create a deep buyer library; a permission request acquisition unit configured to periodically send a permission acquisition request to the buyers in the deep buyer library; a behavior log acquisition unit configured to acquire a behavior log of the buyer in a unit period, wherein the behavior log comprises a number of times that the buyer clicks a recommended commodity, a number of times that the buyer purchases a recommended commodity, a number of times that the buyer marks as not interested, and a number of times that the buyer manually closes a recommendation; an exposure frequency acquisition unit configured to acquire a recommended exposure frequency of a recommended commodity in a unit period; a retention score calculation unit configured to input the behavior log and the recommended exposure frequency into a preset dynamic evaluation model, and output a retention score of the buyer; a retention score determination unit configured to determine whether to remove the buyer from the deep buyer library according to the retention score. 3.The big data based internet mall operation management system according to claim 2, characterized in that, In the retention score determination unit, the content of determining whether to remove the buyer from the deep buyer library according to the retention score comprises: when the retention score is greater than a first preset score threshold, it is determined to keep the buyer in the deep buyer library, and the deep recommendation is continuously provided; when the retention score is not greater than the first preset score threshold but greater than a second preset score threshold, it is determined to keep the buyer in the deep buyer library, and a reminder information is sent to the buyer; when the retention score is not greater than the second preset score threshold, it is determined to remove the buyer.

4. The big data-based Internet mall operation management system according to claim 1, characterized by, The recommendation acceptance determination module comprises: a historical chat information acquisition unit configured to acquire historical chat information of the buyers in the deep buyer library and target friends; a historical mentioned information acquisition unit configured to extract historical mentioned commodity information from the historical chat information; a historical purchase information acquisition unit configured to acquire historical purchase commodity information of the buyers. The recommendation acceptance determining unit is configured to determine the recommendation acceptance of the buyer to each target friend based on the historical mentioned commodity information and the historical purchased commodity information. 5.The big data based internet mall operation management system according to claim 4, characterized in that, The recommendation acceptance determining unit is configured to determine the recommendation acceptance of the buyer to each target friend based on the historical mentioned commodity information and the historical purchased commodity information. comparing the product type and product model in the historical mentioned product information with the product type and product model in the historical purchased product information, if both the product type and product model are the same, determining the matching degree as x 1; if only the product type is the same, determining the matching degree as x 2; if both the product type and product model are not the same, determining the matching degree as x 3; If the historical mentioned commodity mentioned by the target friend is included in the historical purchased commodity purchased by the buyer, an additional matching degree is obtained by summing the matching degrees. A final matching degree is obtained according to the additional matching degree and a preset basic matching degree. The final matching degrees of all target friends are respectively converted into recommendation acceptances of 0-1. 6.The big data based internet mall operation management system according to claim 1, characterized in that, The friend recommendation degree determining module comprises: A recent information obtaining unit is configured to obtain recent chat information of the buyer and the target friend in a monitoring interval in the deep buyer database. A commodity information extracting unit is configured to extract recent mentioned commodity information according to the recent chat information. A mentioned times counting unit is configured to count the mentioned times of each target friend to the recent mentioned commodity. A recommendation degree calculating unit is configured to determine the friend recommendation degree of each recent mentioned commodity according to the mentioned times and the recommendation acceptance of the buyer to each target friend. 7.The big data based internet mall operation management system according to claim 6, characterized in that, In the mentioned times counting unit, when counting the mentioned times of each target friend to the recent mentioned commodity, the chat information containing the recent mentioned commodity is subjected to semantic analysis to determine whether it is valid mentioning. If it is valid mentioning, it is checked whether it is repeated mentioning in the chat information. If not, the mentioned times counting is increased by 1. 8.The big data based internet mall operation management system according to claim 1, wherein, The friend recommendation degree updating module comprises: An information extracting unit is configured to receive search text input in a search box of the mall by the buyer in real time, and extract search commodity information in the search text, including commodity type and commodity model information. A model information determining unit is configured to determine whether the commodity model information exists in the search commodity information. A recommendation degree first updating unit is configured to determine that the friend recommendation degree is unchanged when the commodity model information does not exist in the search commodity information. A same-type commodity extracting unit is configured to extract recent mentioned commodities of the same type based on the commodity type of the search commodity when the commodity model information exists in the search commodity information. A recommendation degree second updating unit is configured to match the commodity model of the search commodity with the commodity model of the recent mentioned commodity of the same type when the recent mentioned commodity of the same type as the search commodity exists, determine whether there is a target recent mentioned commodity successfully matched with the search commodity model, the target recent mentioned commodity refers to the recent mentioned commodity successfully matched with the search commodity model, and update the friend recommendation degree of the target recent mentioned commodity when the target recent mentioned commodity successfully matched with the search commodity model exists. A recommendation degree third updating unit is configured to update the friend recommendation degrees of all recent mentioned commodities of the same type as the search commodity when the target recent mentioned commodity successfully matched with the search commodity model does not exist. A recommendation degree fourth updating unit is configured to determine that the friend recommendation degree is unchanged when the recent mentioned commodity of the same type as the search commodity does not exist. 9.The big data based internet mall operation management system according to claim 8, characterized in that, In the recommendation degree second updating unit, the definition of successful matching is that the similarity between the product models is calculated and compared with a preset similarity threshold, and when the calculated similarity exceeds the preset similarity threshold, it is determined as successful matching. 10.The big data based internet mall operation management system according to claim 1, wherein, The recommendation state matching module comprises: An immediate push determination unit is configured to determine that the recommendation state of the recently mentioned product is an immediate push state when the friend recommendation degree is greater than a first preset recommendation threshold. A recommendation preparation determination unit is configured to determine that the recommendation state of the recently mentioned product is a recommendation preparation state when the friend recommendation degree is greater than a second preset recommendation threshold and is not greater than the first preset threshold. A shield push determination unit is configured to determine that the recommendation state of the recently mentioned product is a shield state when the friend recommendation degree is not greater than the preset second preset recommendation threshold.

Images