Model-based network live broadcast e-commerce intelligent monitoring method and system

By constructing an intelligent monitoring model and utilizing improved image and speech recognition technologies, combined with CRNN and Transformer algorithms, real-time identification and evidence collection of illegal and irregular activities in online live-streaming e-commerce can be achieved. This addresses the shortcomings of traditional monitoring methods and improves regulatory efficiency and fairness.

CN118972665AActive Publication Date: 2024-11-15JIANGSU NET NEW BOCHUANG TECH CO LTD

Patent Information

Application Number
CN202411124633.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-11-15
Estimated Expiration
2044-08-16

AI Technical Summary

Technical Problem

Traditional monitoring methods are insufficient to meet the real-time identification and evidence collection needs of illegal and irregular behaviors in online live-streaming e-commerce. Existing technologies are inadequate in identifying and supervising illegal content and cannot effectively address the real-time, virtual, and cross-regional characteristics of live-streaming e-commerce.

Method used

A reasonable monitoring model is constructed, and intelligent recognition technology is used to analyze live network videos in real time. By combining image text recognition, speech recognition and semantic analysis with improved CRNN algorithm, Hidden Markov model and Transformer algorithm, computer-executable violation matching rules are established to realize intelligent monitoring of illegal and disciplinary violations.

Benefits of technology

It significantly reduces the workload and difficulty of manual supervision, improves the universality of monitoring methods and systems, and can promptly detect and solidify violations, thus maintaining a fair business environment for online e-commerce.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118972665A_ABST
    Figure CN118972665A_ABST
Patent Text Reader

Abstract

The invention discloses a model-based network live broadcast e-commerce intelligent monitoring method and system, and the method comprises the steps: fusing a video image Chinese character recognition method based on an improved CRNN algorithm, a voice recognition method based on a hidden Markov model (HMM) and a Gaussian probability intensity function, and a large language model semantic analysis method based on a Transform algorithm; according to the method, characteristics of illegal and illegal behaviors of network live broadcast are summarized and concluded, a service supervision / monitoring model is constructed, and the illegal and illegal behaviors in the network live broadcast are identified in time through model matching, so that an intelligent and automatic monitoring method and system oriented to network live broadcast e-commerce are provided for market supervision and management departments.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent computing technology, and specifically relates to a model-based online live broadcast e-commerce intelligent monitoring method and system. Background Art

[0002] While the booming development of live e-commerce has brought huge social and economic benefits, it has also brought unprecedented new challenges to government governance and market monitoring / regulation. Cases of illegal, disciplinary and irregular publicity and advertising on online platforms and online operations (the term "violation" is used uniformly in this paper) and consumer rights protection cases have increased year by year. However, due to the real-time, virtual, cross-regional and diverse characteristics of online live broadcasting, as well as its extremely fast transmission speed and wide coverage, traditional monitoring methods are difficult to meet the growing monitoring needs.

[0003] On the other hand, my country's artificial intelligence technologies, including machine vision technology, speech recognition technology, and natural language processing technology, are booming. Intelligent analysis technologies represented by face recognition, vehicle recognition, event monitoring, image retrieval, binocular stereo vision technology, and video splicing technology are widely used in government monitoring and security industries. Speech recognition technology is fully applied in life services, intelligent customer service and other fields. Key technologies such as semantic analysis and large language models are also booming in government affairs and digital economy. However, the monitoring / regulatory applications of technologies such as machine vision text extraction, speech recognition text conversion, semantic analysis, and large language models in the fields of live e-commerce and online advertising have not been fully developed. Problems such as the difficulty in identifying illegal, disciplinary, and regulatory content, the difficulty in obtaining evidence, and the insufficient construction of monitoring / regulatory business models have not been resolved, and cannot meet the growing monitoring needs. Summary of the invention

[0004] In response to the above problems, the present invention constructs a reasonable monitoring model and conducts real-time or quasi-real-time analysis of live network video information to promptly discover or verify illegal, disciplinary and regulatory violations, providing technical and information support for timely disposal.

[0005] To achieve the above purpose, the present invention designs a model-based online live broadcast e-commerce intelligent monitoring method, which monitors online live broadcast behaviors by building a reasonable monitoring model and using intelligent recognition technology. The intelligent monitoring method includes the following steps: S1. Collection of basic data. The basic data includes live broadcast platform data, live broadcast organization data or online e-commerce data, anchor data, complaint and reporting data and / or public opinion monitoring data. The collection of basic data of monitored objects is a basic and common task for intelligent monitoring of online live broadcast e-commerce. S2. Monitoring model construction and maintenance: Based on behaviors expressly prohibited by national or local government laws, regulations, rules and regulations, natural language rule descriptions are converted into mathematical logic expressions to establish a monitoring model that can be automatically executed by a computer, that is, matching rules for illegal, disciplinary and regulatory violations (collectively referred to as violations in this invention) online live broadcast behaviors, and maintain the monitoring model in real time according to new rules or public order and good customs, that is, add new matching rules or modify and improve existing matching rules; The monitoring model includes at least one of an illegal promotion model (B-1), a sales restriction model (B-2), a counterfeit and shoddy model (B-3), an unfair pricing behavior model (B-4, also known as a price violation model), and a false advertising model (B-5); There is no order of precedence between steps S1 and S2; S3. Sampling live network video, the sampling results include video frames and / or audio frames, the sampling method includes random sampling, directional sampling, inspection sampling or daily comprehensive sampling, and the daily comprehensive sampling includes a combination of random sampling and directional sampling; S4. Intelligent recognition, through model matching and calculation of image and text recognition, voice recognition, and semantic analysis in live videos, to identify abnormal behaviors of online live e-commerce; the abnormal behaviors include illegal promotional activities, sales of banned goods, counterfeiting and shoddy products, unfair price competition, false advertising, etc.

[0006] S5. Preserve evidence, including recording and archiving or capturing images, recording the name of the live broadcast room, the address of the live broadcast video, the interception time, the number of people online at the interception time, the illegal content and other information, and archive them to form a record of violations; S6. Clue handling, including recording, summarizing, and displaying violation information, issuing warning signals, and reminding manual handling.

[0007] Furthermore, the live broadcast platform data includes unified social credit code, enterprise name, platform name, etc.; The live broadcasting organization data includes unified social credit code, company name, etc.; The online e-commerce data include unified social credit code, enterprise name, online business registration information, online store address, compulsory certification product name, certification information, etc., as well as geographical indication product information, geographical indication product certification information, etc.; The anchor data includes the name of the live broadcast platform, live broadcast room information, anchor personal information (including name, ID number, permanent residence, etc.), etc.; The complaint report data includes the webcast address information of the complained party, the complaint content information, etc. The public opinion monitoring data includes live broadcast room information, public opinion content information, etc. The monitoring model also includes an unfairness model and / or an abnormal order model.

[0008] Furthermore, the illegal promotion model includes at least one of the following models: (1) The formula for identifying food marketing with no bottom line for minors is: (live broadcast promotional audio and video ∃ illegal words ∈ minors’ characteristic word sample package) ∩ (live broadcast promotional audio and video ∃ illegal words ∈ spoof and vulgar characteristic word sample package), where ∃ means existence, ∈ means belonging, and ∩ means intersection (referring to the data set, the same below); (2) Using the formula for identifying major political events and activities: (live broadcast promotional audio and video ∃ illegal words ∈ major event feature word sample package) ∨ (product image LOGO = official logo of major event); where ∨ represents a union; (3) Identification formula for marketing that violates public order and good morals: live broadcast promotional audio and video ∃violation word ∈ {x|x is a public order and good morals characteristic word}; (4) Violation of the seven-day unconditional return and exchange marketing identification formula: (live broadcast promotion audio and video ∃ illegal word ∈ {x|x is the seven-day unconditional return and exchange feature word}) ∨ (live broadcast promotion audio and video ∃ illegal word ∈ {x|x is the no-refund and no-exchange feature word}); (5) The formula for identifying marketing violations with the final right of interpretation is: live broadcast promotional audio and video ∃violation word ∈ {x|x is a characteristic word of the right of interpretation}.

[0009] Furthermore, the sales restriction model includes at least one of the following models: (1) The sales behavior identification formula of products that do not meet the national standards and industry standards for protecting human health and personal and property safety is: Live broadcast product quality standard ∉{x|x is a national mandatory standard, a national recommended standard, or an industry standard}, where ∉ means it does not belong; (2) Adulteration or counterfeiting of products, passing off fake products as genuine ones, passing off inferior products as good ones, or passing off substandard products as qualified ones. Identification formula: (product LOGO = brand registered trademark) ∩ (actual product selling price < official brand price of the same product * k 1 ); where the price comparison coefficient k for the same product 1 The value range is between 10% and 50%; (3) The formula for identifying sales behaviors of products that have been explicitly eliminated by the state is: (live broadcast promotional audio and video ∃ illegal word ∈ {x|x is a characteristic word for prohibited or restricted sales products}) ∨ (live broadcast promotional audio and video ∃ illegal word ∈ {x|x is a characteristic word for eliminated products}); (4) The sales behavior identification formula for products with falsified origin, products with falsified or counterfeit factory names and addresses, and products with falsified or counterfeit certification marks and other quality marks is as follows: (product advertised origin area ∉ mark approval area) ∨ (actual sales products ≠ certification mark registered products).

[0010] Furthermore, the counterfeit and shoddy model includes at least one of the following models: (1) Formula for identifying the fictitious origin of goods protected by geographical indications: the advertised origin of the goods ∉ the area where the geographical indication is approved; (2) The formula for identifying the act of selling goods that infringe the exclusive rights of a registered trademark is: (product LOGO ≈ brand registered trademark) ∩ (product category ∈ registered trademark approved category).

[0011] Furthermore, the unfair pricing behavior model (also called the price violation model) includes at least one of the following models: (1) The core formula for e-commerce live streaming price violations: The deviation between the live streaming price and the fair price F = (current price - fair price) / fair price. The formula is used to further determine whether it is price gouging, a reasonable price, or unfair competition using a monopoly position; (2) Data determination of live broadcast "special price, wholesale price, ex-factory price, market lowest price": the data volume of original price and comparison price = 0; (3) Live limited-time promotion, limited-time special offer: limited-time period data = 0; (4) Buy one get one free live broadcast with free gifts: Free gift data = 0; (5) Improper formula for marking original prices in live broadcasts: actual price of the product < original price & sales data for the past seven days = 0; (6) Live broadcast promotion formula: The formula for raising the price first and then raising the price again is: live event price ≥ platform re-purchase price > platform first purchase price; The price increase formula is: platform repurchase price > live event price > platform first purchase price; The formula for price increase after price reduction is: first purchase price > re-purchase price and promotion price > re-purchase price; False price reduction formula: repurchase price > promotion price = first purchase price; False discount formula: repurchase price > first purchase price > promotion price; Formula for lottery sales: gift price ≥ k 2 , where k 2 Price threshold for gifts.

[0012] Furthermore, the false advertising model includes at least one of the following models: (1) Formula for identifying fictitious origin: the origin data claimed in the live broadcast > the known origin data of the product; (2) Formula for identifying fictitious performance: performance data promoted in live broadcast > known performance standards of the product; (3) Formula for identifying fictitious uses: Live broadcast product use promotion > known product uses or government-approved certification content; (4) Formula for identifying fictitious honors: the honors claimed by the live broadcast product > the relevant honor database data; (5) False promotion identification formula, live broadcast false "special price, wholesale price, factory price, market lowest price" formula: comparison price data volume = 0, or live broadcast product price ≥ comparison price data; (6) Live broadcast of false limited-time promotions or limited-time special offers: the limited-time data = 0, or the limited-time data > the promotional period; (7) Formula for identifying fictitious original prices: the original price claimed in the live broadcast > the historical price of the same product on the e-commerce platform; (8) Formula for identifying fictitious licenses: Live broadcast claims of approved products > administrative license database data; (9) Patent fictitious identification formula: patent content claimed in live broadcast > patent database data.

[0013] Furthermore, the random sampling includes capturing a certain amount of video frames from past videos or real-time videos according to random rules; the targeted sampling includes searching for matching video sources for sampling based on market or public complaint information or public opinion monitoring abnormal information; the inspection sampling includes spot-checking a certain amount or all video frame information of each subject in sequence according to the registered live broadcast platform, live broadcast organization, online e-commerce or anchor information.

[0014] Further, step S4 includes: S41. Chinese character recognition in video images, including a Chinese character recognition method in video images based on an improved CRNN algorithm; S42. Speech recognition, including speech recognition methods based on hidden Markov models and Gaussian probability intensity functions; S43, semantic analysis, performing semantic analysis on the text recognition results and / or speech recognition results in the video image, wherein the semantic analysis method includes a large language model semantic analysis method based on the Transformer algorithm, using a word joint embedding vector to create a connection between words that share a word; adding an Attention structure at the word vector level (the Attention structure in deep learning mainly refers to the attention mechanism, which allows the model to dynamically focus on the information part that is most relevant to the current task when processing sequence data. This mechanism has been widely used in the fields of natural language processing and machine translation), while capturing the word granularity contextual relationship, adding the word granularity contextual relationship; after encoding each fused Token, performing Pooling-Mean, and then connecting the classifier for prediction; S44, model matching and calculation (C-4), matching the semantic analysis result with the model constructed in step S2, or matching the corresponding data set constructed by the model, or conditionally matching the training evaluation function constructed by the model, so as to calculate the corresponding matching result, that is, to give a conclusion on whether there is any violation and which rule is violated.

[0015] Furthermore, the evidence storage in step S5 includes the record storage of the violating subject, violating keywords and violating matching model information; the violating subject includes at least one of a live broadcast organization, an online e-commerce company and an anchor.

[0016] Further, the intelligent monitoring method includes a method for identifying similar illegal, disciplinary and regulatory violations based on dynamic association (which can be recorded as step S5a); the dynamic association includes an analysis of whether each new violation record is associated with a past violation record, and generally only the same type of violation records (referring to the same violation record matching model) are subjected to association analysis, and different types of violation records do not need to be associated; the objects of the dynamic association include at least one of the violation subject association, video source association, commodity association, and video content association, and the video content association includes at least one of the image matching association, text matching association, and voice matching association. The commodity association refers to the same commodity or essentially the same commodity or associated commodities sold in a package, etc. The associated commodities sold in a package include the lack of any one of the commodities or several related commodities, and the role of other commodities cannot be played or reflected; the calculation technology of the matching association is all existing technology, and the present invention uses the correlation coefficient (or correlation degree) to characterize the strength of the association characteristic; The method for identifying similar illegal, disciplinary and regulatory violations includes a suspected similar identification method (i.e. suspected similar illegal, disciplinary and regulatory violations), wherein the suspected similar identification method includes at least one correlation coefficient of the violating subject, the involved commodity (based on the commodity association calculation result) or the video source is not less than the third threshold K. 3 , K 3 ≥90%, and at least one of the correlation coefficients of image matching association, text matching association, and voice matching association is not less than the fourth threshold K 4 , K 4 ≥80%.

[0017] Furthermore, the method for identifying similar illegal, disciplinary and regulatory violations also includes a method for confirming similar identification, wherein the product of the correlation coefficients of at least two of the image matching association, the text matching association and the voice matching association is greater than the fifth matching threshold K 5 , or the product of the three correlation coefficients is greater than the sixth matching threshold K 6 , K 5 >K 6At this time, it is judged that the two videos from different sources are closely related, that is, they belong to the same type of illegal, disciplinary and regulatory violations. Of course, the same-source video has the highest matching degree, and it is not meaningful to do correlation analysis because they are actually the same thing.

[0018] On the other hand, a model-based online live e-commerce intelligent monitoring system, the system includes a sampling module, a model set and an evidence storage module respectively connected to the intelligent recognition module, and a clue handling module connected to the evidence storage module, the sampling module is respectively connected to the external input module and the sampling rule module; the intelligent recognition module includes a video slicing unit, a picture and text recognition unit, a semantic recognition unit and a model matching unit respectively connected in sequence, and another branch voice recognition unit, a text extraction unit, a semantic recognition unit and a model matching unit connected in sequence, wherein the semantic recognition unit and the model matching unit are shared; the clue handling module includes a clue preliminary review unit, a transfer supervision unit and a handling result unit connected in sequence; the model set includes at least one of an illegal promotion model, a prohibited and restricted sales model, a counterfeit and shoddy model, an unfair pricing behavior model, and a false propaganda model; the external input module includes at least one of live broadcast platform data, live broadcast organization data, online e-commerce data, and anchor data, and also includes complaint and reporting data and / or public opinion monitoring data.

[0019] Furthermore, the monitoring system implements any one of the above-mentioned model-based intelligent monitoring methods for live online e-commerce.

[0020] The advantages and beneficial effects of the present invention are as follows: by integrating a Chinese character recognition method for video images based on an improved CRNN algorithm, a speech recognition method based on a hidden Markov model (HMM) and a Gaussian probability intensity function, and a large language model semantic analysis method based on a Transformer algorithm, and matching them with a newly constructed business supervision / monitoring model, illegal, disciplinary, and regulatory violations can be identified, and the problems of difficulty in identifying illegal content and obtaining evidence can be solved, thereby significantly reducing the workload and difficulty of manual supervision; at the same time, by designing and applying an obviously unfair model and an abnormal ordering model, the universality of the monitoring method and system is improved, and a fair operating environment for online e-commerce is further reasonably maintained; and the distributed architecture design and productized functional module design can adapt to regulatory needs of different scales and scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a schematic diagram of the model-based live e-commerce intelligent monitoring method and system principle flow chart; Figure 2 This is a schematic diagram of an example of live video image and speech text conversion and semantic analysis. DETAILED DESCRIPTION

[0022] The specific implementation of the present invention is further described below in conjunction with the accompanying drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention. Example 1

[0023] like Figure 1 As shown, the present invention designs a model-based online live broadcast e-commerce intelligent monitoring method, which realizes the monitoring of online live broadcast behavior by constructing a reasonable monitoring model and using intelligent recognition technology. The intelligent monitoring method includes the following steps: S1. Collection of basic data. The basic data includes live broadcast platform data, live broadcast organization data or online e-commerce data, anchor data, complaint and reporting data, and / or public opinion monitoring data. The collection of basic data of monitored objects is a basic and common task for intelligent monitoring of online live broadcast e-commerce. The basic data collection includes data source management and data governance; The data source management includes configuring the national online transaction supervision platform (needs to be connected to the online transaction platform of the State Administration for Market Regulation), the data sources of the online transaction supervision platforms of various provinces and cities (based on the actual configuration of each province), configuring the national 12315 complaint reporting system (needs to be connected to the 12315 complaint reporting system of the State Administration for Market Regulation), the data sources of the 12345 platforms of various provinces and cities (based on the actual configuration of each province), and configuring the data sources of the public opinion monitoring systems of various provinces and cities; configuring the live broadcast address of the online live broadcast room (dynamically configured according to complaints and reports and public opinion data); The data governance includes data cleaning, data comparison and data association based on the collected data source data. Data governance technologies are all existing technologies.

[0024] S2. Monitoring model construction and maintenance: Based on the behaviors expressly prohibited by the laws, regulations, rules and regulations of the national or local government, the rule descriptions in natural language are converted into mathematical logic expressions, and a monitoring model that can be automatically executed by a computer is established, that is, the matching rules for illegal, disciplinary and illegal (collectively referred to as violations in this invention) online live broadcast behaviors. The monitoring model can be maintained in real time according to new rules or public order and good customs, that is, new matching rules are added or existing matching rules are modified and improved; The monitoring models include at least one of the illegal promotion model (B-1), the prohibited and restricted sales model (B-2), the counterfeit and shoddy product model (B-3), the unfair pricing behavior model (B-4, also known as the price violation model), the false advertising model (B-5), etc.; the naming of different models is simply based on the customary naming of illegal, disciplinary and regulatory violations. In fact, different naming may overlap with other behaviors, that is, a certain behavior may simultaneously meet the identification rules of two or more models.

[0025] The present invention converts the traditional natural language description of illegal, disciplinary and regulatory violations into a strict mathematical logic expression, thereby making it possible to transform manual analysis and judgment into direct computer analysis and judgment, which is a prerequisite for realizing intelligent monitoring. The proposal of such a model and the subsequent model building method are both the first of the present invention. The model architecture is very convenient for the later maintenance of the monitoring system because it is closer to the expression form of natural language and is more analyzed and judged from the perspective of established rules or habitual cognition. The existing other live data monitoring methods are more based on the results of learning to analyze and judge, which requires enough learning cases, and the learning cases generally require manual intervention, and inform the learning system of its high-probability judgment results, so as to improve the accuracy of system monitoring, and its adaptability to new rules is not strong. Every time a new rule appears, the system needs to have enough cases to learn. Compared with the model building method of the present invention that only needs to convert the rules into logical expressions, the workload and complexity of the present invention are obviously much lower, that is, the system is more convenient to understand and maintain.

[0026] There is no order of precedence between steps S1 and S2; S3. Sampling live network video, the sampling results include video frames and / or audio frames, the sampling methods include random sampling, directional sampling, patrol sampling or daily comprehensive sampling, the daily comprehensive sampling includes a combination of random sampling and directional sampling; this embodiment adopts a daily comprehensive sampling method that combines random sampling with directional sampling.

[0027] S4. Intelligent recognition, through model matching and calculation of image and text recognition, voice recognition, and semantic analysis in live videos, to identify abnormal behaviors of online live e-commerce; the abnormal behaviors include illegal promotional activities, sales of banned goods, counterfeiting and shoddy products, unfair price competition, false advertising, etc.

[0028] This embodiment mainly integrates the text recognition method in video images based on the improved CRNN algorithm, the speech recognition method based on the hidden Markov model (HMM) and Gaussian probability intensity function, and the large language model semantic analysis method based on the Transformer algorithm to perform video live content recognition and model calculation.

[0029] S5. Preserve evidence, including recording and archiving or capturing images, and form a record of violations; Live videos suspected of illegal, disciplinary, or regulatory violations are recorded and archived or downloaded or captured as images. The violation records also include the violating subject, violation keywords, violation matching model, video source, network flow information, download volume, click volume, and other related information to consolidate evidence. This embodiment mainly records the name of the live broadcast room, the address of the live video, the interception time, the number of people online at the interception time, the illegal content, and other information, and archives them.

[0030] S6. Clue handling, including recording, summarizing, and displaying violation information, issuing warning signals, and reminding manual handling.

[0031] According to the jurisdiction of the supervision object, the clues will be classified and assigned at different levels, and warnings, guidance, interviews and investigations will be given according to the law. The disposal results will be fed back to the data source department in real time and recorded in the file.

[0032] Preferably, the live broadcast platform data includes a unified social credit code, company name, platform name, etc.; The live broadcasting organization data includes unified social credit code, company name, etc.; The online e-commerce data include unified social credit code, enterprise name, online business registration information, online store address, compulsory certification product name, certification information, etc., as well as geographical indication product information, geographical indication product certification information, etc.; The anchor data includes the name of the live broadcast platform, live broadcast room information, anchor personal information (including name, ID number, permanent residence, etc.), etc.; The complaint report data includes the webcast address information of the complained party, the complaint content information, etc. The public opinion monitoring data includes live broadcast room information, public opinion content information, etc. The monitoring model also includes an unfairness model and / or an abnormal order model. The unfairness model formula includes: ((actual selling price of the product <n 1 * the cost price of the product) & (order quantity ≥ order lower limit n 2 ) & (no time limit for the activity or no limit on the number of purchases)), and / or ((cost price - actual selling price) * order volume ≥ the platform profit of the merchant throughout the year); the abnormal order model formula includes: (unit time interval t 1 The total number of orders within ≥ the upper limit of orders), the upper limit of orders can be set = n 3 *Daily average daily order volume, such as (total number of orders within 1 minute ≥ 10 * daily average daily order volume), or (total number of orders within 10 minutes ≥ 30 * daily average daily order volume), or (total number of orders within 60 minutes ≥ 60 * daily average daily order volume), or a combination of the above formulas, etc.; where n 1 is the apparent unfairness coefficient. In this embodiment, n is set 1 =0.1, n 2 is the lower limit of the order, in this embodiment, n 2 =5,n 3 is the order limit coefficient, t 1 is the order quantity statistics interval, as shown in the above example t 1 、n 3Different combinations can be adopted, and specific system maintenance personnel can limit them according to factors such as regional income level, consumption capacity or commodity category.

[0033] The obviously unfair model and the abnormal order model are set in this embodiment mainly to deal with the possible large losses caused by the merchant's promotional errors that exceed their tolerance, or the losses are obviously too large and far exceed the punishment that should be borne for the mistakes, that is, it is obviously unfair. The monitoring system needs to be able to discover and provide evidence in time to maintain social fairness. This is the first of its kind in the present invention, and the current existing monitoring system has neither such function nor effective analysis and judgment models and methods.

[0034] This embodiment constructs a series of business supervision / monitoring models by summarizing and generalizing the characteristics of current illegal, disciplinary and regulatory violations in online live broadcasts.

[0035] Preferably, the illegal promotion model (B-1, program design number, the same below) includes at least one of the following models: (1) The formula for identifying food marketing with no bottom line for minors is: (live broadcast promotional audio and video ∃ illegal words ∈ minors’ characteristic words sample package) ∩ (live broadcast promotional audio and video ∃ illegal words ∈ spoof and vulgar characteristic words sample package), where ∃ means existence, ∈ means belonging to, and ∩ means intersection (referring to the data set, the same below); the minors’ characteristic words can be updated dynamically; (2) Using the formula for identifying major political events and activities: (live broadcast promotional audio and video ∃ illegal words ∈ major event feature word sample package) ∨ (product image LOGO = official logo of major event); where ∨ represents a union; the major event feature words can be dynamically updated; (3) Identification formula for marketing that violates public order and good morals: live broadcast promotional audio and video ∃violation word ∈ {x|x is a public order and good morals characteristic word}; the public order and good morals characteristic word can be dynamically updated; (4) Violation of the seven-day no-reason return and exchange marketing identification formula: (live broadcast promotion audio and video ∃ violation word ∈ {x|x is a seven-day no-reason return and exchange feature word}) ∨ (live broadcast promotion audio and video ∃ violation word ∈ {x|x is a no-refund and no-exchange feature word}); the seven-day no-reason return and exchange feature words include no returns once sold, no returns, no deposit refunds, pre-sale products are not eligible for seven-day no-reason return, etc., and no-refund and no-exchange feature words include no returns and no exchanges, no refunds, no returns, no three guarantees for prizes and gifts, etc., and the feature words can be updated dynamically; (5) Identification formula for marketing with final right of interpretation that violates regulations: live broadcast promotional audio and video ∃violation word ∈ {x|x is a characteristic word of right of interpretation}; the characteristic words of right of interpretation include that the final right of interpretation belongs to XX, etc.

[0036] Preferably, the sales restriction model (B-2) includes at least one of the following models: (1) The sales behavior identification formula of products that do not meet the national standards and industry standards for protecting human health and personal and property safety is: Live broadcast product quality standard ∉{x|x is a national mandatory standard, a national recommended standard, or an industry standard}, where ∉ means it does not belong; (2) Adulteration or counterfeiting of products, passing off fake products as genuine ones, passing off inferior products as good ones, or passing off substandard products as qualified ones. Identification formula: (product LOGO = brand registered trademark) ∩ (actual product selling price < official brand price of the same product * k 1 ); where the price comparison coefficient k for the same product 1 The value range is between 10% and 50%. 1 =20%; (3) The formula for identifying sales behaviors of products that have been explicitly eliminated by the state is: (live broadcast promotional audio and video ∃ illegal words ∈ {x|x is a characteristic word for prohibited or restricted products}) ∨ (live broadcast promotional audio and video ∃ illegal words ∈ {x|x is a characteristic word for eliminated products}); characteristic words for eliminated products include clearance sales, company bankruptcy clearance sales, clearance specials, etc.; characteristic words can be updated dynamically; (4) The sales behavior identification formula for products with falsified origin, products with falsified or counterfeit factory names and addresses, and products with falsified or counterfeit certification marks and other quality marks is as follows: (product advertised origin area ∉ mark approval area) ∨ (actual sales products ≠ certification mark registered products).

[0037] Preferably, the counterfeit and shoddy model (B-3) includes at least one of the following models: (1) Formula for identifying the fictitious origin of goods protected by geographical indications: the advertised origin of the goods ∉ the area where the geographical indication is approved; (2) The formula for identifying the act of selling goods that infringe the exclusive rights of a registered trademark is: (product LOGO ≈ brand registered trademark) ∩ (product category ∈ registered trademark approved category); the symbol “≈” can be judged by methods such as image comparison with a matching degree exceeding 90% or 95%.

[0038] Preferably, the unfair pricing behavior model (B-4, also known as the price violation model) includes at least one of the following models: (1) The core formula for e-commerce live streaming price violations: The deviation between the live streaming price and the fair price F = (current price - fair price) / fair price. The formula is used to further determine whether it is price gouging, a reasonable price, or unfair competition using a monopoly position. In this embodiment, if F ≥ 3, it is determined to be price gouging, and if F ≤ 0.2, it is determined to be unfair competition using a monopoly position. (2) Data determination of live broadcast "special price, wholesale price, ex-factory price, market lowest price": the data volume of original price and comparison price = 0; (3) Live limited-time promotion, limited-time special offer: limited-time period data = 0; (4) Buy one get one free live broadcast with free gifts: Free gift data = 0; (5) Improper formula for marking original prices in live broadcasts: actual price of the product < original price & sales data for the past seven days = 0; (6) Live broadcast promotion formula: The formula for price increase by first raising and then raising is: live broadcast event price ≥ platform re-purchase price > platform first purchase price; platform first purchase price refers to the normal sales price before the event, the event price is the actual sales price collected during the event, and the platform re-purchase price refers to the commodity price collected again from the online store after the event ends; the so-called first raising and then raising means that merchants first raise the sales price tag before holding a live broadcast sales event, and then raise the price again during the event through bidding and other methods, in order to achieve the purpose of abnormal price increase and profit; The price increase formula is: platform re-purchase price > live event price > platform first purchase price; that is, the product price is increased before the event, and the price is reduced during and after the event; The formula for price reduction first and then increase is: first purchase price > re-purchase price and event price > re-purchase price; that is, the product price is reduced before the event, and the price is increased during and after the event; False price reduction formula: repurchase price > event price = first purchase price; that is, the price before and during the event is the same, and the price increases after the event; False discount formula: repurchase price > first purchase price > promotion price; that is, after the marketing activity is over, the price of the product is increased to achieve the purpose of false discount; Formula for lottery sales: gift price ≥ k 2 , where k 2 is the gift price threshold. In this embodiment, the gift price threshold k is set 2 =50,000 yuan.

[0039] Among them, items (2) to (6) are the formulas for deriving prices that violate regulations in e-commerce live streaming.

[0040] Preferably, the false advertising model (B-5) includes at least one of the following models: (1) Formula for identifying fictitious origin: the origin data claimed in the live broadcast > the known origin data of the product; (2) Formula for identifying fictitious performance: performance data promoted in live broadcast > known performance standards of the product; (3) Formula for identifying fictitious uses: Live broadcast product use promotion > known product uses or government-approved certification content; (4) Formula for identifying fictitious honors: the honors claimed by the live broadcast product > the relevant honor database data; (5) False promotion identification formula, live broadcast false "special price, wholesale price, factory price, market lowest price" formula: comparison price data volume = 0, or live broadcast product price ≥ comparison price data; (6) Live broadcast of false limited-time promotions or limited-time special offers: the limited-time data = 0, or the limited-time data > the promotional period; (7) Formula for identifying fictitious original prices: the original price claimed in the live broadcast > the historical price of the same product on the e-commerce platform; (8) Formula for identifying fictitious licenses: Live broadcast claims of approved products > administrative license database data; (9) Patent fictitious identification formula: patent content claimed in live broadcast > patent database data.

[0041] Preferably, the random sampling includes grabbing a certain amount of video frames from past videos or real-time videos according to random rules; the directional sampling includes searching for matching video sources for sampling according to market or public complaint and reporting information or public opinion monitoring abnormal information; the inspection sampling includes spot-checking a certain amount or all video frame information of each subject in order according to the registered live broadcast platform, live broadcast organization, online e-commerce or anchor information, such as spot-checking or full inspection in alphabetical order; the random sampling, directional sampling and daily comprehensive sampling are essentially an active monitoring sampling method, while directional sampling is a passive monitoring sampling method; in actual applications, the ideal situation is to monitor all live videos, that is, to conduct full inspection sampling and analysis. Considering that the existing computing power is difficult to achieve real-time sampling and analysis of all online live broadcast behaviors, the necessity of full inspection is not very great, and monitoring is mainly carried out by sampling; in actual engineering practice, a method combining active monitoring random sampling with passive monitoring sampling is usually adopted, that is, daily sampling monitoring according to certain rules and immediate directional sampling monitoring when there is a report.

[0042] Preferably, step S4 comprises: S41. Chinese character recognition in video images (C-1, program design number, the same below), including Chinese character recognition method in video images based on improved CRNN algorithm; Whether it is live broadcast monitoring or advertising monitoring, text recognition in the video must be completed.

[0043] For text recognition in video graphics based on the CRNN (Convolution Recurrent Neural Network) algorithm, please refer to the paper "An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition", which proposes a new neural network architecture that integrates feature extraction, sequence modeling, and transcription into a unified framework. It is tested on standard benchmark datasets including IIIT-5K, Street View Text, and ICDAR datasets, and is state-of-the-art.

[0044] This embodiment adopts a method for recognizing text in video images based on an improved CRNN algorithm for video text recognition: Specifically, the video stream is processed into frames of pictures through OpenCV technology, and then a series of texts are detected using a target detection model. For the detected text, the text is scaled to a uniform size and normalized. Based on the principle of convolutional neural networks, the stride parameter is optimized and adjusted to solve the problem of large differences and dispersion in the width of Chinese characters, which is prone to text recognition errors, and to achieve accurate extraction of data features. Through recurrent neural networks, the traditional CRNN (Convolution Recurrent Neural Network) algorithm is improved to a bidirectional LSTM (Long Short-Term Memory, a commonly used recurrent neural network Recurrent Neural Network, RNN) algorithm to solve the problems of gradient vanishing and long-term dependence in RNN in traditional algorithms.

[0045] S42. Speech recognition (C-2), including speech recognition methods based on hidden Markov models and Gaussian probability intensity functions; Live broadcasts or advertising videos contain a lot of voice information. In live broadcasts, the host constantly incites the user's emotions to achieve the effect of live broadcast sales or user rewards. Therefore, voice recognition technology occupies an important position in this video monitoring. Generally, a voice recognition system is divided into three stages: signal processing stage, sound stage and decoding stage.

[0046] This embodiment adopts a speech recognition method based on a hidden Markov model (HMM) and a Gaussian probability intensity function: Signal preprocessing: Noise suppression and noise removal are performed on speech signals. Valid speech segments of speech signals are identified through methods such as energy analysis and zero-crossing rate detection. First, the acoustic waveform in the video is divided into audio fragment frames (usually 10ms, 15ms or 20ms), and the audio fragment frames are converted into spectral features.

[0047] Acoustic Modeling: The Hidden Markov Model (HMM) is used to describe the relationship between the hidden state sequence and the observable acoustic feature sequence in the speech signal. The Hidden Markov Model is a probabilistic model of time series, which describes the process of randomly generating an unobservable state sequence (StateSequence) by a hidden Markov chain, and then generating an observation sequence (Observation Sequence) by generating an observation from each state. Each position in the sequence can be regarded as a moment. In this information processing stage, it can be used to model acoustic observations (feature vectors) at the subword level (such as Chinese phonemes). Usually three states are modeled for each phoneme, respectively for the beginning, middle and end of the phoneme. Each state has a self-transition and a transition to the next state.

[0048] The commonly used method at this stage is to calculate the probability intensity function in a continuous space. The acoustic signal of the speech is transformed into a series of feature vectors through a series of signal processing transformations, each of which represents a time slice of the speech input. This embodiment uses a statistical technique such as the Gaussian probability intensity function to perform the calculation of the above process and try to identify a single sound.

[0049] Language modeling: Language modeling deals with the vocabulary and grammatical structure issues in speech recognition. By building a language model, the rationality of different vocabulary sequences can be evaluated, thereby guiding the decoding process to select recognition results that are more in line with language habits. In the recognition stage, the feature sequence of the input speech signal is matched with the established HMM template, and metrics such as likelihood or distance are calculated.

[0050] Decoding and output: Using the word pronunciation dictionary and language model, the VITERBI algorithm is used to determine the word sequence with the highest probability for a given acoustic event. First, the Viterbi table is initialized, and a two-dimensional table is used to store the intermediate results, where the rows of the table correspond to different time steps (or observations) and the columns correspond to different hidden states. The speech signal to be recognized is then converted into a series of observations, which constitute the input sequence for the algorithm to process. Starting from the second column, the language model is traversed for each slice, the cumulative probability is calculated, and the best matching result is retained as the final state, mapped to the corresponding output symbol sequence (usually a word or phoneme in the vocabulary), and finally output as the recognition result of the speech recognition system.

[0051] S43, semantic analysis (C-3), semantic analysis is performed on the text recognition results and / or speech recognition results in the video image, the semantic analysis method includes a large language model semantic analysis method based on the Transformer algorithm, based on the use of traditional word vectors and / or word vectors, word joint embedding vectors are used at the same time to create connections between words that share words; an Attention structure is added at the word vector level (the Attention structure mainly refers to the attention mechanism in deep learning, which allows the model to dynamically focus on the information part that is most relevant to the current task when processing sequence data, This mechanism has been widely used in natural language processing, machine translation and other fields. While capturing the character-level contextual relationship, the word-level contextual relationship is added. The establishment of the contextual relationship here is not only for the traditional adjacent contextual relationship, but the system first performs a correlation test within the same limited time video. When there are obviously related characters or words (such as a correlation coefficient greater than 90%), the character-level contextual relationship and / or word-level contextual relationship across time intervals is captured and established; after encoding each Token after the word fusion, mean pooling (Pooling-Mean) is performed, and then the classifier is used for prediction; the present invention improves the semantic analysis method of the large language model based on the Transformer algorithm, which is more suitable for the phenomenon of short tone and frequent semantic jumps (that is, the semantics of adjacent discourses may not be closely related, and the semantics after a period of time may be closely related) in the current online live broadcast. This is one of the innovative points of the present invention.

[0052] The range of videos in live webcasts is wide, and the models for different scenarios are different. It is necessary to quickly build an effective model for screening. Especially for some sudden illegal advertising buzzwords, the speed of model production is particularly critical. Compared with other conventional algorithms, the model trained by the Transformer algorithm can be trained using distributed GPUs, which can improve training efficiency. In addition, the live video is mainly spoken by the host, and the correlation between the previous and the next may not be very strong. Sometimes the topic is talked about for a long time and then comes back to introduce it. It is necessary to capture semantics with a longer interval for association. In this regard, that is, when analyzing and predicting longer texts, the Transformer-based algorithm is better at capturing semantic associations with a longer interval.

[0053] Therefore, this embodiment selects the Transformer algorithm for semantic analysis, such as Figure 2 shown.

[0054] Text preprocessing: Convert raw text data into a format suitable for model processing, including removing irrelevant characters, stop words, and noise data through data cleaning, dividing the text into smaller units through word segmentation, reducing vocabulary diversity by restoring word forms to basic formats, and achieving more extensive word form merging through stemming.

[0055] CLS tag addition: Based on the CLS (Concept Learning System) specification, each token is formed by fusing the words of the text recognition results and / or speech recognition results.

[0056] Text vectorization: Through word embedding technology, words or sentences are converted into vector representations to capture their semantic information. Each embedding is then encoded (encoder), and the encoding results form several related sequences (real-time feature words).

[0057] Feature extraction: At the word vector level, the Attention structure (model library structure, connotation model feature words) is added to capture the contextual relationship at the character level while adding the contextual relationship at the word level. Then, a classifier (SoftmaxClassifier, a general classifier) ​​is used for classification to form a semantic expression that is easy for computers to recognize.

[0058] S44, model matching and calculation (C-4), the extracted features are matched with the model constructed in step S2, or the data set corresponding to the model is matched, or the training evaluation function of the model is matched, so as to calculate the corresponding matching results, that is, to give the conclusion whether there is any violation and which rule is violated. The specific matching method is as follows: Figure 2 As shown, the encoding results are formed into several related sequences and the model feature words in the Attention structure are matched with the model (the ⊕ symbol in the figure represents the model matching operation). This embodiment adopts the deep semantic matching model DSSM (Deep Structured Semantic Models) method, and the extracted feature vector is used as the query to calculate the cosine similarity with the key word vectors of the model library constructed in B-1, B-2, B-3, B-4, and B-5. If the chord value is closer to 1, that is, the angle tends to 0°, the greater the similarity of the two vectors; the closer the cosine value is to 0, that is, the closer the angle tends to 90°, the less similar the two vectors are. This embodiment sets the chord value to be higher than 0.8 as a basic match, and the record is used as a clue for evidence.

[0059] Preferably, the evidence storage in step S5 includes record storage of the violating subject, violating keywords and violating matching model information; the violating subject includes at least one of a live broadcast organization, an online e-commerce company and an anchor. Example 2

[0060] The difference from Example 1 is that the intelligent monitoring method described in this embodiment also includes a method for identifying similar illegal, disciplinary and regulatory violations based on dynamic association (which can be recorded as step S5a). The similar illegal, disciplinary and regulatory violations include repeated or multiple similar illegal, disciplinary and regulatory violations, such as one person exaggerating the same product in different occasions, illegally promoting, or selling the same restricted product on different platforms, etc., which is equivalent to multiple and repeated violations, and the punishment should be increased, but the premise is that the monitoring system needs to grasp accurate information in a timely manner; the dynamic association includes an analysis of whether each new violation record is associated with a past violation record, and generally only the same type of violation records ( The dynamic association object includes at least one of the association of the violating subject, the association of the video source, the association of the commodity, and the association of the video content. The association of the video content includes at least one of the association of the image matching, the association of the text matching, and the association of the voice matching. The commodity association refers to the same commodity or essentially the same commodity or associated commodities sold in a package, etc. The associated commodities sold in a package include the lack of any one of the commodities or several related commodities, and the role of other commodities cannot be played or reflected. The matching association technologies are all existing technologies. The present invention uses the correlation coefficient (or correlation degree) to characterize the strength of the association characteristics. The method for identifying similar illegal, disciplinary and regulatory violations includes a suspected similar identification method (i.e. suspected similar illegal, disciplinary and regulatory violations), wherein the suspected similar identification method includes at least one correlation coefficient of the violating subject, the involved commodity (based on the commodity association calculation result) or the video source is not less than the third threshold K. 3 , K 3 ≥90%, and at least one of the correlation coefficients of image matching association, text matching association, and voice matching association is not less than the fourth threshold K 4 , K 4 ≥80%. In this embodiment, K 3 =98%, K 4 =90%; where K is set 3 =98% correlation coefficient requirement, mainly considering repeated violations, at least one of the anchor, live broadcast platform, online e-commerce, live broadcast organization, and product is the same or essentially the same; set K 4=90% correlation coefficient requirement is mainly to consider that the video content association does not need to be completely or almost the same in terms of image, text, voice, etc., because the forms of expressing the same meaning in Chinese may vary greatly. Therefore, a requirement lower than the third threshold is set. However, considering that only one of the image, text, and voice is required, a higher standard is also reasonable. That is, if the correlation coefficients of the three items of image, text, and voice are not high enough, it is reasonable to consider that they are not related. Therefore, this embodiment sets K 4 =90%.

[0061] Preferably, the method for identifying similar illegal, disciplinary and regulatory violations also includes a method for confirming similar identification, wherein the product of the correlation coefficients of at least two of the image matching association, the text matching association and the voice matching association is greater than the fifth matching threshold K 5 , or the product of the three correlation coefficients is greater than the sixth matching threshold K 6 , K 5 >K 6 Considering that any correlation coefficient of image content association should not be less than 60%, and at least one correlation coefficient should not be less than 90%, we are more conservative and generally take K 5 ≥0.6, K 6 ≥0.4, in this embodiment, K 5 =0.8, K 6 =0.6; of course K 5 , K 6 The larger the two parameter values ​​are, the higher the credibility of the identification of similar illegal, disciplinary and regulatory violations will be, but this will also increase the probability of missed identification. If the two parameter values ​​are relatively small, the recognition success rate will increase (that is, the missed identification rate will decrease), but the probability of misidentification or false alarm will also increase, which needs to be determined by the user based on the actual situation. At this time, it is judged that the two videos from different sources are closely related, that is, they belong to the same type of illegal, disciplinary and regulatory violations. Of course, the same-source videos have the highest matching degree, and there is little meaning in doing correlation analysis because they are actually the same thing.

[0062] Although the calculation method of the correlation coefficient can adopt a relevance calculation method similar to that of a general search engine, establishing different correlation judgment criteria is still the first of its kind in the present invention. It converts some of people's ideas into strict mathematical expressions, which is convenient for computer execution, making the monitoring of online live e-commerce more intelligent and the warning information more timely and effective.

[0063] The identification of similar illegal, disciplinary and regulatory violations is mainly to provide a basis and information for the handling and punishment of illegal behaviors, and to enable timely discovery and timely handling of multiple violators and behaviors; the present invention not only preserves evidence but also records information such as the number of downloads or views, the type of downloaders or viewers, the sales volume of related products, the transaction price, the consumer groups, etc., and then analyzes the degree of network impact, the affected population, the amount of illegal profits, etc., which also provides a basis and information for handling and punishment. Example 3

[0064] The difference from Example 1 is that this embodiment only performs directional sampling, that is, when there is a report or abnormal public opinion, the directional video is sampled. Example 4

[0065] A model-based online live e-commerce intelligent monitoring system, the system includes a sampling module, a model set and an evidence storage module respectively connected to an intelligent recognition module, and a clue handling module connected to the evidence storage module, the sampling module is respectively connected to an external input module and a sampling rule module; the intelligent recognition module includes a video slicing unit, a picture and text recognition unit, a semantic recognition unit and a model matching unit respectively connected in sequence, and another branch voice recognition unit, a text extraction unit, a semantic recognition unit and a model matching unit connected in sequence, wherein the semantic recognition unit and the model matching unit are shared; the clue handling module includes a clue preliminary review unit, a transfer supervision unit and a handling result unit connected in sequence; the model set includes at least one of an illegal promotion model, a prohibited and restricted sales model, a counterfeit and shoddy model, an unfair pricing behavior model, and a false propaganda model; the external input module includes at least one of live broadcast platform data, live broadcast organization data, online e-commerce data, and anchor data, and also includes complaint and reporting data and / or public opinion monitoring data.

[0066] Preferably, the model-based online live e-commerce intelligent monitoring system implements at least one of Examples 1 to 3 and their preferred methods.

[0067] Constructing a suitable supervision / monitoring model is the key to the implementation of the present invention. It is necessary to combine various laws, regulations, rules and public order and good customs to summarize the characteristics of current illegal, disciplinary and irregular behaviors in online live broadcasts to form an operation model that can be recognized by computers. Moreover, as society develops and the situation changes, the monitoring model can be maintained in real time according to new rules or public order and good customs, that is, new matching rules can be added or existing matching rules can be modified and improved, so that the system has higher practical value.

[0068] The basic principle of the model-based online live e-commerce intelligent monitoring method designed by the present invention is to form a monitoring model that can be recognized by a computer by reasonably summarizing and generalizing the characteristics of current online live illegal, disciplinary and regulatory violations, and utilize a video image Chinese character recognition method based on an improved CRNN algorithm, a speech recognition method based on a hidden Markov model (HMM) and a Gaussian probability intensity function, and a large language model semantic analysis method based on a Transformer algorithm to parse and convert the real-time online, dynamically transformed video images and video speech of the live e-commerce into text that can be formatted and stored, perform semantic analysis through a large language model, and match the business monitoring model to identify illegal, disciplinary and regulatory violations, thereby solving the problems of difficulty in identifying illegal, disciplinary and regulatory content and difficulty in obtaining evidence for the content, and significantly reducing the workload and difficulty of manual supervision. At the same time, through distributed architecture design and productized functional design, it can adapt to monitoring / regulatory needs of different scales and scenarios.

[0069] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the technical principles of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the protection scope of the present invention. For example, the number and combination of monitoring models can form several different embodiment combinations, and the third to sixth threshold setting selection in the dynamic association-based similar illegal, disciplinary and regulatory behavior identification method can also form several new embodiment combinations, and so on.

Claims

1. A model-based intelligent monitoring method for live e-commerce. It is characterized in that By building a monitoring model and using intelligent recognition technology to monitor live broadcasting behavior on the Internet, the intelligent monitoring method includes the following steps: S1. Collection of basic data, including live broadcast platform data, live broadcast organization data or online e-commerce data, anchor data, complaint and reporting data and / or public opinion monitoring data; S2. Construction and maintenance of monitoring models. Based on the behaviors expressly prohibited by national or local government laws, regulations, rules and regulations, the rules description in natural language is converted into mathematical logic expressions to establish a monitoring model for automatic execution by computers; The monitoring model includes at least one of an illegal promotion model, a sales restriction model, a counterfeit and shoddy model, an unfair pricing behavior model, and a false advertising model; S3. Sampling live webcast videos, wherein the sampling method includes random sampling, directional sampling, inspection sampling or daily comprehensive sampling, wherein the daily comprehensive sampling includes a combination of random sampling and directional sampling; S4, intelligent recognition, through model matching and calculation of image and text recognition, voice recognition, and semantic analysis in live video, to identify abnormal behaviors of online live e-commerce; S5. Preserve evidence, including recording and archiving or capturing images, and form a record of violations; S6. Clue handling, including recording, summarizing, and displaying violation information, issuing warning signals, and reminding manual handling; There is no order of precedence between steps S1 and S2; Step S4 includes: S41. Chinese character recognition in video images, including a Chinese character recognition method in video images based on an improved CRNN algorithm; S42. Speech recognition, including speech recognition methods based on hidden Markov models and Gaussian probability intensity functions; S43, semantic analysis, performing semantic analysis on the text recognition results and / or speech recognition results in the video image, wherein the semantic analysis method includes a large language model semantic analysis method based on the Transformer algorithm, using word joint embedding vectors to create connections between words that share characters; adding an Attention structure at the word vector level to capture the word granularity context relationship while adding the word granularity context relationship; after encoding each Token after word fusion, performing mean pooling, and then connecting to a classifier for prediction; S44, model matching and calculation, matching the semantic analysis result with the model constructed in step S2 for relevance, thereby calculating the corresponding matching result.

2. According to the model-based online live e-commerce intelligent monitoring method of claim 1, It is characterized in that The live broadcast platform data includes unified social credit code, enterprise name, and platform name; The live broadcasting organization data includes unified social credit code and company name; The online e-commerce data includes unified social credit code, enterprise name, online business registration information, online store address, compulsory certification product name, and certification information; The anchor data includes the name of the live broadcast platform, live broadcast room information, and anchor personal information; The complaint report data includes the webcast address information and complaint content information of the complained party; The public opinion monitoring data includes live broadcast room information and public opinion content information; The monitoring model also includes an unfairness model and / or an abnormal order model.

3. According to the model-based online live e-commerce intelligent monitoring method of claim 1, It is characterized in that The illegal promotion model includes at least one of the following models: (1) The formula for identifying food marketing with no bottom line for minors is: (live broadcast promotional audio and video ∃ illegal words ∈ minors’ characteristic word sample package) ∩ (live broadcast promotional audio and video ∃ illegal words ∈ spoof and vulgar characteristic word sample package), where ∃ means existence, ∈ means belonging, and ∩ means intersection; (2) Using the formula for identifying major political events and activities: (live broadcast promotional audio and video ∃ illegal words ∈ major event feature word sample package) ∨ (product image LOGO = official logo of major event); where ∨ represents a union; (3) Identification formula for marketing that violates public order and good morals: live broadcast promotional audio and video ∃violation word ∈ {x|x is a public order and good morals characteristic word}; (4) Violation of the seven-day unconditional return and exchange marketing identification formula: (live broadcast promotion audio and video ∃ illegal word ∈ {x|x is the seven-day unconditional return and exchange feature word}) ∨ (live broadcast promotion audio and video ∃ illegal word ∈ {x|x is the no-refund and no-exchange feature word}); (5) The formula for identifying marketing violations with the final right of interpretation is: live broadcast promotional audio and video ∃violation word ∈ {x|x is a characteristic word of the right of interpretation}.

4. According to the model-based online live e-commerce intelligent monitoring method of claim 3, It is characterized in that The sales restriction model includes at least one of the following models: (1) The sales behavior identification formula of products that do not meet the national standards and industry standards for protecting human health and personal and property safety is: Live broadcast product quality standard ∉{x|x is a national mandatory standard, a national recommended standard, or an industry standard}, where ∉ means it does not belong; (2) Adulteration or counterfeiting of products, passing off fake products as genuine ones, passing off inferior products as good ones, or passing off substandard products as qualified ones. Identification formula: (product LOGO = brand registered trademark) ∩ (actual product selling price < official brand price of the same product * k 1 ); where the price comparison coefficient k for the same product 1 The value range is between 10% and 50%; (3) The formula for identifying sales behaviors of products that have been explicitly eliminated by the state is: (live broadcast promotional audio and video ∃ illegal word ∈ {x|x is a characteristic word for prohibited or restricted sales products}) ∨ (live broadcast promotional audio and video ∃ illegal word ∈ {x|x is a characteristic word for eliminated products}); (4) The sales behavior identification formula for products with falsified origin, products with falsified or counterfeit factory names and addresses, and products with falsified or counterfeit certification marks and other quality marks is as follows: (product advertised origin area ∉ mark approval area) ∨ (actual sales products ≠ certification mark registered products).

5. According to the model-based online live e-commerce intelligent monitoring method of claim 1, It is characterized in that The counterfeit and shoddy models include at least one of the following models: (1) Formula for identifying the fictitious origin of goods protected by geographical indications: the advertised origin of the goods ∉ the area where the geographical indication is approved; (2) The formula for identifying the act of selling goods that infringe the exclusive rights of a registered trademark is: (product LOGO ≈ brand registered trademark) ∩ (product category ∈ registered trademark approved category).

6. According to the model-based online live e-commerce intelligent monitoring method of claim 1, It is characterized in that The improper price behavior model includes at least one of the following models: (1) The core formula for e-commerce live streaming price violations: The deviation between the live streaming price and the fair price F = (current price - fair price) / fair price. The formula is used to further determine whether it is price gouging, a reasonable price, or unfair competition using a monopoly position; (2) Live broadcast "special price, wholesale price, factory price, market lowest price" data determination: original price, comparison price data volume = 0; (3) Live limited-time promotion, limited-time special offer: limited-time period data = 0; (4) Buy one get one free live broadcast with free gifts: Free gift data = 0; (5) Improper formula for marking original prices in live broadcasts: actual price of the product < original price & sales data for the past seven days = 0; (6) Live broadcast promotion formula: The formula for raising the price first and then raising the price again is: live event price ≥ platform re-purchase price > platform first purchase price; The price increase formula is: platform repurchase price > live event price > platform first purchase price; The formula for price increase after price reduction is: first purchase price > re-purchase price and promotion price > re-purchase price; False price reduction formula: repurchase price > promotion price = first purchase price; False discount formula: repurchase price > first purchase price > promotion price; Formula for lottery sales: gift price ≥ k 2 , where k 2 Price threshold for gifts.

7. According to the model-based online live e-commerce intelligent monitoring method of claim 1, It is characterized in that The false advertising model includes at least one of the following models: (1) Formula for identifying fictitious origin: the origin data claimed in the live broadcast > the known origin data of the product; (2) Formula for identifying fictitious performance: performance data promoted in live broadcast > known performance standards of the product; (3) Formula for identifying fictitious uses: Live broadcast product use promotion > known product uses or government-approved certification content; (4) Formula for identifying fictitious honors: the honors claimed by the live broadcast product > the relevant honor database data; (5) False promotion identification formula, live broadcast false "special price, wholesale price, factory price, market lowest price" formula: comparison price data volume = 0, or live broadcast product price ≥ comparison price data; (6) Live broadcast of false limited-time promotions or limited-time special offers: the limited-time data = 0, or the limited-time data > the promotional period; (7) Formula for identifying fictitious original prices: the original price claimed in the live broadcast > the historical price of the same product on the e-commerce platform; (8) Formula for identifying fictitious licenses: Live broadcast claims of approved products > administrative license database data; (9) Patent fictitious identification formula: patent content claimed in live broadcast > patent database data.

8. According to the model-based online live e-commerce intelligent monitoring method of claim 1, It is characterized in that The random sampling includes capturing a certain number of video frames from past videos or real-time videos according to random rules; the targeted sampling includes searching for matching video sources for sampling based on market or public complaint information or public opinion monitoring abnormal information; the inspection sampling includes spot-checking a certain amount of video frame information for each subject in sequence according to the registered live broadcast platform, live broadcast organization, online e-commerce or anchor information.

9. According to the model-based online live e-commerce intelligent monitoring method of claim 1, It is characterized in that The evidence storage in step S5 includes the record storage of the violating subject, violating keywords and violating matching model information; the violating subject includes at least one of the live broadcast organization, online e-commerce and anchor.

10. According to the model-based online live e-commerce intelligent monitoring method of claim 1, It is characterized in that The intelligent monitoring method includes a method for identifying similar illegal, disciplinary and regulatory violations based on dynamic association, wherein the dynamic association includes, for each new violation record, an analysis of whether it is associated with a previous violation record; the object of the dynamic association includes at least one of violation subject association, video source association, commodity association and video content association, wherein the video content association includes at least one of image matching association, text matching association and voice matching association; The method for identifying similar illegal, disciplinary and regulatory violations includes a suspected similar identification method, wherein at least one of the correlation coefficients of the violating subject, the commodity or the video source is not less than the third threshold K. 3 , K 3 ≥90%, and at least one of the correlation coefficients of image matching association, text matching association, and voice matching association is not less than the fourth threshold K 4 , K 4 ≥80%.

11. According to claim 10, a model-based online live e-commerce intelligent monitoring method, It is characterized in that The method for identifying similar illegal, disciplinary and regulatory violations also includes a method for confirming similar identification, wherein the product of the correlation coefficients of at least two items among image matching association, text matching association and voice matching association is greater than the fifth matching threshold K. 5 , or the product of the three correlation coefficients is greater than the sixth matching threshold K 6 , K 5 >K 6 At this time, it is judged that the two videos from different sources belong to the same type of illegal, disciplinary and regulatory violations.

12. A model-based online live e-commerce intelligent monitoring system. It is characterized in that The monitoring system implements the model-based online live e-commerce intelligent monitoring method described in any one of claims 1 to 11, and the system includes a sampling module, a model set and a evidence storage module respectively connected to the intelligent recognition module, and a clue handling module connected to the evidence storage module, and the sampling module is respectively connected to the external input module and the sampling rule module; the intelligent recognition module includes a video slicing unit, a picture and text recognition unit, a semantic recognition unit and a model matching unit respectively connected in sequence, and another branch voice recognition unit, a text extraction unit, a semantic recognition unit and a model matching unit connected in sequence, wherein the semantic recognition unit and the model matching unit are shared; the clue handling module includes a clue preliminary review unit, a transfer supervision unit and a handling result unit connected in sequence; the model set includes at least one of an illegal promotion model, a prohibited and restricted sales model, a counterfeit and shoddy model, an unfair pricing behavior model, and a false propaganda model; the external input module includes at least one of live broadcast platform data, live broadcast organization data, online e-commerce data, and anchor data, and also includes complaint and reporting data and / or public opinion monitoring data.

Citation Information

Patent Citations

  • user speech semantic analysis method for a network live broadcast scene

    CN109657241A

  • Chinese sentence semantic intelligent matching method and device based on multi-granularity fusion model

    CN111310438A

  • System and method for compliance quality inspection and interactive question answering of live broadcast scene

    CN114786035A

  • Anchor live broadcast behavior auditing method and device, electronic equipment and readable storage medium

    CN115393929A

  • Live broadcast sensitive information monitoring method and system

    CN117641043A

Cited By

  • Multi-modal real-time compliance auditing method, system and device for investment adviser live broadcast and storage medium

    CN120951137A