Method and device for identifying false propaganda of live broadcast commodity selling based on consumer comments

Abstract

The application discloses a live broadcast goods false propaganda identification method based on consumer comments and related devices. The method comprises the following steps: obtaining consumer comment text and preprocessing; performing semantic coding on the preprocessed consumer comment text to obtain a comment semantic representation; inverting the comment semantic representation to obtain an attribute semantic component and a violation semantic component; generating a potential propaganda commitment anchor point based on the attribute semantic component; calculating the commitment violation degree based on the potential propaganda commitment anchor point and the violation semantic component; and outputting the identification result of whether the current consumer comment involves live broadcast goods false propaganda according to the commitment violation degree. The application can improve the identification accuracy, pertinence and applicability of live broadcast goods false propaganda comments even in the absence of live broadcast original propaganda content.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and related apparatus for identifying false advertising in live-streaming e-commerce based on consumer reviews. Background Technology

[0002] With the rapid development of live-streaming e-commerce, live-streaming sales have become an important form of online retail. Hosts influence consumers' purchasing decisions through real-time explanations, product demonstrations, and interactive marketing. However, problems such as exaggerated claims, vague promises, false efficacy claims, and discrepancies between advertised and actual products are becoming increasingly prominent, posing challenges to consumer rights protection and platform governance. Therefore, effectively identifying false advertising in live-streaming sales has become an urgent technical problem to be solved. Current technologies for identifying false advertising largely rely on analyzing advertising copy, product detail pages, live-streaming videos, speech-to-text transcripts, or manually reported information. These methods typically require direct access to the live-streaming promotional content. However, the real-time nature, flexible expression, and rapid changes in live-streaming scripts make it difficult to completely collect and accurately model the promotional content, thus limiting the effectiveness of related methods in live-streaming sales scenarios.

[0003] Consumer reviews, feedback from consumers after purchasing and using products, can reflect discrepancies between the actual product experience and the advertised claims during livestreaming. Therefore, using consumer reviews to identify false advertising has significant application value. However, existing review-based analysis methods mostly focus on sentiment analysis, complaint identification, or anomaly detection, making it difficult to effectively distinguish between ordinary negative reviews and those involving false advertising. Furthermore, consumer reviews typically do not directly contain the original advertising content from the livestream, making it difficult to establish a direct correlation between advertised promises and actual experience, thus affecting the accuracy of identification. Therefore, there is an urgent need for a new method for identifying false advertising in livestreaming e-commerce, capable of effectively identifying false advertising based on consumer reviews even in the absence of the original livestream advertising content. Summary of the Invention

[0004] To address the problems of existing technologies where consumer reviews are often implicitly expressed, easily confused with ordinary negative reviews, and lack a direct correlation with live-stream promotional content, leading to insufficient accuracy in identifying false advertising, this invention proposes a method and related apparatus for identifying false advertising in live-stream e-commerce based on consumer reviews. This method does not rely directly on the original live-stream script or promotional text; it can extract implicit mismatches in promotional promises from consumer reviews, effectively reducing the interference of ordinary negative reviews on the identification results and improving the accuracy, relevance, and applicability of identifying false advertising reviews in live-stream e-commerce. It is applicable to various scenarios such as live-stream e-commerce platforms, product review analysis systems, platform risk warning systems, and consumer rights protection systems.

[0005] To achieve the above objectives, the technical solution of the present invention is as follows:

[0006] The method for identifying false advertising in live-streaming e-commerce based on consumer reviews includes the following steps:

[0007] Obtain consumer review text and preprocess it; then perform semantic encoding on the preprocessed consumer review text to obtain the semantic representation of the review;

[0008] Invert the semantic representation of the comment to obtain the semantic components of attributes and the semantic violation components;

[0009] Generate potential advertising commitment anchors based on attribute semantic components;

[0010] The degree of commitment violation is calculated based on potential propaganda commitment anchors and semantic components that violate the commitment.

[0011] Based on the degree of breach of promise, the system outputs the identification result of whether the current consumer reviews involve false advertising in live-streaming e-commerce.

[0012] Preferably, the preprocessing includes special character removal, stop word removal, and fixed input length.

[0013] Preferably, a BERT pre-trained language model is used for semantic encoding.

[0014] Preferably, the semantic representation of the comment is inverted to obtain the attribute semantic components and the violation semantic components, including the following steps:

[0015] The semantic representation of the comment is projected into the attribute semantic space and the violation semantic space respectively to obtain the attribute semantic vector and the violation semantic vector.

[0016] Calculate the attention distribution of attribute semantic vectors and violation semantic vectors. Based on the attention distribution calculation results, aggregate the attribute semantic vector sequence and the violation semantic vector sequence to obtain the attribute semantic component and the violation semantic component.

[0017] Preferably, generating potential advocacy commitment anchors based on attribute semantic components includes the following steps:

[0018] Construct a commitment anchor space, which includes commitment anchors corresponding to different commitment modes;

[0019] The similarity between the semantic components of the attributes and the commitment anchors is calculated, and the similarity is normalized to obtain the weight of each commitment anchor.

[0020] Based on the weight of each commitment anchor, potential promotional commitment anchors corresponding to the current consumer review are generated.

[0021] Preferably, calculating the degree of commitment breach includes the following steps:

[0022] Characteristics of commitment violation relationships are determined based on the differences and interactions between potential propaganda commitment anchors and semantic components of violation;

[0023] The commitment violation relationship features are input into the violation degree calculation module to obtain the initial commitment violation degree. :

[0024]

[0025] in, Indicates the degree of initial commitment breach. Let r be the transformation matrix, and r be the characteristic of the commitment violation relation. For bias, For activation functions;

[0026] Calculate violation triggering gating based on potential advocacy commitment anchors and violation semantic components. The formula is as follows:

[0027]

[0028] The initial commitment violation degree is corrected based on the violation trigger gating to obtain the commitment violation degree. The formula is as follows:

[0029] .

[0030] Preferably, based on the degree of promise breach, the system outputs an identification result regarding whether the current consumer review involves false advertising in live-streaming e-commerce, including the following steps:

[0031] Preset recognition threshold;

[0032] If the degree of breach of promise in the current consumer review exceeds the identification threshold, then the current consumer review involves false advertising in live-streaming e-commerce; if the degree of breach of promise in the current consumer review does not exceed the identification threshold, then the current consumer review does not involve false advertising in live-streaming e-commerce.

[0033] Based on the above, the present invention also discloses a device for identifying false advertising in live-streaming e-commerce based on consumer reviews, comprising:

[0034] The acquisition module is used to acquire and preprocess consumer review text;

[0035] The identification module is used to semantically encode the preprocessed consumer review text to obtain the semantic representation of the review; invert the semantic representation of the review to obtain attribute semantic components and violation semantic components; generate potential advertising commitment anchors based on the attribute semantic components; calculate the commitment violation degree based on the potential advertising commitment anchors and violation semantic components; and output the identification result of whether the current consumer review involves false advertising in live streaming e-commerce based on the commitment violation degree.

[0036] Based on the above, the present invention also discloses a computer device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement any of the methods described above.

[0037] Based on the foregoing, the present invention also discloses a readable storage medium storing a computer program, which, when executed by a processor, implements any of the methods described above.

[0038] Based on the above technical solution, the beneficial effects of the present invention are:

[0039] 1) This invention does not rely on the complete acquisition of live video, the original script of the anchor or the product promotion text, but can identify false advertising in live sales based solely on consumer comments. Therefore, it can reduce the difficulty of data collection and application costs, and improve the applicability of the method in real live streaming scenarios.

[0040] 2) This invention performs semantic inversion on consumer reviews, distinguishing and representing attribute semantics and violation semantics in the reviews. This reduces the interference of ordinary negative emotions, general complaints, or after-sales dissatisfaction on the results of false advertising identification, thereby improving the targeting and accuracy of identification.

[0041] 3) This invention generates potential advertising commitment anchors, and constructs a referable advertising commitment representation for consumer reviews in the absence of original live broadcast advertising content, thereby establishing an indirect correspondence between advertising commitments and actual experiences, solving the problem that existing methods are difficult to form an effective comparison;

[0042] 4) By calculating the degree of commitment violation between potential advertising commitment anchors and the semantic components of the violation, this invention achieves explicit modeling of the core feature of "unfulfilled advertising commitments," which can more effectively identify false advertising clues hidden in consumer reviews.

[0043] 5) The technical solution of this invention has a clear structure and is easy to implement. It can be widely applied to scenarios such as product risk screening, intelligent comment analysis, platform content governance, and consumer rights protection on live e-commerce platforms, and has good practical value. Attached Figure Description

[0044] Figure 1 This is a schematic flowchart of a method for identifying false advertising in live-streaming e-commerce based on consumer reviews in one embodiment.

[0045] Figure 2 This is a flowchart of a method for identifying false advertising in live-streaming e-commerce based on consumer reviews, as exemplified in one embodiment. Detailed Implementation

[0046] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0047] like Figure 1 , 2 As shown, this embodiment provides a method for identifying false advertising in live-streaming e-commerce based on consumer reviews, including the following steps:

[0048] Step 1: Obtain consumer review text and preprocess it; perform semantic encoding on the preprocessed consumer review text to obtain the semantic representation of the review.

[0049] 1.1 Special Character Removal: Remove invalid symbols, duplicate spaces, and abnormal characters from comment text.

[0050] 1.2 Stop word removal: The Harbin Institute of Technology Chinese stop word list is used to remove stop words that do not contribute to the recognition results.

[0051] 1.3 Fixed Input Length: The length of the comment is truncated or padded to meet the requirements of subsequent model processing. Specifically, the fixed input length is 64 Chinese characters; any shortfall is padded with zeros, while any excess is truncated.

[0052] 1.4 Input the preprocessed consumer review text into the BERT pre-trained language model to obtain the corresponding semantic representation of the review:

[0053]

[0054] in, This represents the context semantic vector corresponding to the i-th character.

[0055] Step 2: Invert the semantic representation of the comment to obtain the attribute semantic components and the violation semantic components.

[0056] Because consumer reviews typically include descriptions of product attributes, expressions of user experience, emotional complaints, and indirect accusations of false advertising, directly classifying them based on the overall semantic meaning can easily misclassify ordinary negative reviews as false advertising reviews. Therefore, this invention performs semantic inversion on the semantic representation of the review context, decomposing it into attribute semantic components and violation semantic components.

[0057] in:

[0058] 1) The attribute semantic components are used to represent the product promotion attribute information involved in the review;

[0059] 2) The violated semantic component is used to characterize semantic information in the comments regarding unfulfilled, inconsistent, exaggerated, or distorted claims.

[0060] 2.1 Perform attribute semantic space projection and violation semantic space projection on the semantic representation of the comment context respectively:

[0061]

[0062]

[0063] in, Represents an attribute semantic vector. This indicates a violation of semantic vectors. , , and These are learnable parameters.

[0064] 2.2 Calculate the attention distribution of attribute semantic vectors and violation semantic vectors:

[0065]

[0066]

[0067] in, For self-attention mechanism, The attention score represents the value of the i-th character in the attribute semantic vector. This represents the attention score for violating the semantic vector for the i-th character.

[0068] 2.3 The attribute semantic vector sequence and the violation semantic vector sequence are aggregated separately to obtain the attribute semantic components represented at the sentence level. and violate semantic components :

[0069]

[0070]

[0071] Through the above-mentioned semantic inversion steps, the expression of product attributes and the distortion of advertising in consumer reviews can be distinguished and modeled, thereby reducing the interference of ordinary emotional complaints on subsequent identification results.

[0072] Step 3: Generate potential advocacy commitment anchors based on attribute semantic components.

[0073] Since consumer reviews typically do not directly contain the original promotional content from the live stream, it is difficult to establish a one-to-one correspondence between reviews and the streamer's promotional rhetoric. To address this issue, this invention introduces potential promotional commitment anchors, which are used to construct a corresponding promotional commitment reference for the current review in the absence of the original promotional text from the live stream.

[0074] 3.1 Constructing the Commitment Anchor Space:

[0075]

[0076] in, Let represent the k-th commitment anchor vector. The commitment anchor space is used to characterize common promotional commitment patterns in live-streaming e-commerce scenarios, such as efficacy commitments, material commitments, and specification commitments. Instead of manually writing the text content for each commitment anchor, the model can automatically learn and form anchor representations corresponding to different commitment patterns through training.

[0077] 3.2 Integrate the attribute semantic components from 2.3 Similarity calculation with the commitment anchor:

[0078]

[0079] in, This indicates the similarity between the semantic components of the current comment's attributes and the k-th commitment anchor. .

[0080] 3.3 Normalize the similarity scores to obtain the weights of each commitment anchor point:

[0081]

[0082] 3.4 Generate a representation of the potential promotional commitment anchor for the current comment based on the weight of each commitment anchor:

[0083]

[0084] in, This represents the potential advertising promise anchor points derived from consumer reviews.

[0085] This step allows for the creation of a corresponding promotional commitment reference for comments, even when the original promotional text of the live stream cannot be directly obtained, providing a basis for subsequent calculation of commitment breach.

[0086] Step 4: Calculate the degree of commitment violation based on potential propaganda commitment anchors and semantic violation components.

[0087] After obtaining the potential promotional commitment anchor in step 3.4 And the semantically violated components in step 2.3 Then, further analysis is conducted to determine whether there is a relationship of unfulfilled promises, inconsistent advertising, or exaggerated advertising between the two parties, and the degree of breach of promise is calculated.

[0088] 4.1 First, construct the commitment violation relationship feature r between potential promotional commitment anchors and the semantic component that violates the commitment. The relationship feature may include difference features and interaction features.

[0089]

[0090] in, This indicates the distinguishing features between potential propaganda promise anchors and semantically contradictory components. This represents the element-wise interaction features between potential propaganda promise anchors and semantically contradictory components.

[0091] 4.2 Input the commitment violation relationship features into the violation degree calculation module to obtain the initial commitment violation degree:

[0092]

[0093] in, Indicates the degree of initial commitment breach. The transformation matrix is... For bias, is the activation function. The promise violation degree is used to characterize the degree of mismatch between the potential promotional promises involved in the current consumer review and the actual consumer feedback. The higher the promise violation degree, the more likely the review involves false advertising.

[0094] 4.3 Calculate violation trigger gating This further filters out comments that do not fall under the category of false advertising, such as ordinary complaints, negative logistics reviews, and customer service dissatisfaction.

[0095]

[0096] 4.4 Correcting the initial commitment violation degree based on the violation trigger gating to obtain the commitment violation degree. :

[0097]

[0098] By setting violation trigger gates, the ability of this invention to distinguish between ordinary negative comments and false advertising comments can be further improved.

[0099] Step 5: Output the results of the false advertising identification.

[0100] Based on the degree of breach of promise Output the result of whether the current consumer review involves false advertising in live-streaming e-commerce. Specifically, it is expressed as:

[0101]

[0102] in, The comments indicate that the live-streaming e-commerce involved false advertising. This indicates that the comment does not involve false advertising related to live-streaming e-commerce. The preset identification threshold in this embodiment is 0.5, but to adapt to the needs of different live-streaming scenarios, operators are allowed to fine-tune it within the range of 0.4 to 0.7 based on feedback from the actual live-streaming situation.

[0103] It should be understood that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowchart above may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0104] Based on the same inventive concept, this application also provides an apparatus for implementing the aforementioned method for identifying false advertising in live-streaming e-commerce based on consumer reviews. The solution provided by this apparatus is similar to the solution described in the above method, and therefore will not be repeated here.

[0105] In one embodiment, a device for identifying false advertising in live-streaming e-commerce based on consumer reviews is also provided, including:

[0106] The acquisition module is used to acquire and preprocess consumer review text;

[0107] The identification module is used to semantically encode the preprocessed consumer review text to obtain the semantic representation of the review; invert the semantic representation of the review to obtain attribute semantic components and violation semantic components; generate potential advertising commitment anchors based on the attribute semantic components; calculate the commitment violation degree based on the potential advertising commitment anchors and violation semantic components; and output the identification result of whether the current consumer review involves false advertising in live streaming e-commerce based on the commitment violation degree.

[0108] In the above embodiments, the modules of the live-streaming e-commerce false advertising identification device based on consumer reviews can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0109] In one embodiment, a computer device is also provided, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps as described in all the above method embodiments.

[0110] In one embodiment, a readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps as described in all the above method embodiments.

[0111] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0112] The embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0113] The above are merely preferred embodiments of the present application and are not intended to limit the embodiments of the present application. For those skilled in the art, the embodiments of the present application can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of the present application should be included within the protection scope of the embodiments of the present application.

Claims

1. A method for identifying false advertising in live-streaming e-commerce based on consumer reviews, characterized in that, Includes the following steps: Obtain consumer review text and preprocess it; then perform semantic encoding on the preprocessed consumer review text to obtain the semantic representation of the review; Invert the semantic representation of the comment to obtain the semantic components of attributes and the semantic violation components; Generate potential advertising commitment anchors based on attribute semantic components; The degree of commitment violation is calculated based on potential propaganda commitment anchors and semantic components that violate the commitment. Based on the degree of breach of promise, the system outputs the identification result of whether the current consumer reviews involve false advertising in live-streaming e-commerce.

2. The method for identifying false advertising in live-streaming e-commerce based on consumer reviews according to claim 1, characterized in that, The preprocessing includes special character removal, stop word removal, and fixed input length.

3. The method for identifying false advertising in live-streaming e-commerce based on consumer reviews according to claim 1, characterized in that, Semantic encoding is performed using the BERT pre-trained language model.

4. The method for identifying false advertising in live-streaming e-commerce based on consumer reviews according to claim 1, characterized in that, The process of inverting the semantic representation of comments to obtain attribute semantic components and violation semantic components includes the following steps: The semantic representation of the comment is projected into the attribute semantic space and the violation semantic space respectively to obtain the attribute semantic vector and the violation semantic vector. Calculate the attention distribution of attribute semantic vectors and violation semantic vectors. Based on the calculation results of the attention distribution, aggregate the attribute semantic vector sequence and the violation semantic vector sequence to obtain the attribute semantic component and the violation semantic component.

5. The method for identifying false advertising in live-streaming e-commerce based on consumer reviews according to claim 1, characterized in that, Generating potential advocacy commitment anchors based on attribute semantic components includes the following steps: Construct a commitment anchor space, which includes commitment anchors corresponding to different commitment modes; The similarity between the semantic components of the attributes and the commitment anchors is calculated, and the similarity is normalized to obtain the weight of each commitment anchor. Based on the weight of each commitment anchor, potential promotional commitment anchors corresponding to the current consumer review are generated.

6. The method for identifying false advertising in live-streaming e-commerce based on consumer reviews according to claim 1, characterized in that, Calculating the degree of commitment breach includes the following steps: Characteristics of commitment breach relationships are determined based on the differences and interactions between potential propaganda commitment anchors and breach semantic components; The commitment violation relationship features are input into the violation degree calculation module to obtain the initial commitment violation degree. : ； in, Indicates the degree of initial commitment breach. Let r be the transformation matrix, and r be the characteristic of the commitment violation relation. For bias, For activation functions; Calculate violation triggering gating based on potential advocacy commitment anchors and violation semantic components. The formula is as follows: ； The initial commitment violation degree is corrected based on the violation trigger gating to obtain the commitment violation degree. The formula is as follows: 。 7. The method for identifying false advertising in live-streaming e-commerce based on consumer reviews according to claim 1, characterized in that, Based on the degree of breach of promise, the system outputs the identification result of whether the current consumer review involves false advertising in live-streaming e-commerce, including the following steps: Preset recognition threshold; If the degree of breach of promise corresponding to the current consumer review is greater than the identification threshold, then the current consumer review involves false advertising in live-streaming e-commerce. If the degree of breach of promise corresponding to the current consumer review is not greater than the identification threshold, then the current consumer review does not involve false advertising in live-streaming e-commerce.

8. A device for identifying false advertising in live-streaming e-commerce based on consumer reviews, characterized in that, include: The acquisition module is used to acquire and preprocess consumer review text; The recognition module is used to semantically encode the preprocessed consumer review text to obtain a semantic representation of the review; Invert the semantic representation of comments to obtain attribute semantic components and violation semantic components; generate potential propaganda commitment anchors based on attribute semantic components; The degree of commitment violation is calculated based on the potential advertising commitment anchor and the semantic component of the violation; based on the degree of commitment violation, the result of identification of whether the current consumer review involves false advertising in live streaming e-commerce is output.

9. A computer device, characterized in that, Includes: memory, used to store computer programs; A processor for executing the computer program to implement the method as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images