Method and device for identifying commodity, electronic equipment and storage medium

Abstract

The present disclosure provides a commodity identification method and device, electronic equipment and storage medium, relates to the field of artificial intelligence, in particular to the technical field of large models, natural language processing, and the like, including: acquiring multi-modal data in a live broadcast process, performing semantic analysis on the multi-modal data to obtain structured semantic data; acquiring structured claim data for a live broadcast commodity in the live broadcast process according to the structured semantic data; acquiring factual information of the live broadcast commodity; identifying the live broadcast commodity according to the structured semantic data, the structured claim data and the factual information to obtain an identification result of the live broadcast commodity.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, specifically to the fields of natural language processing and large models, and particularly to a method, apparatus, electronic device and storage medium for identifying goods. Background Technology

[0002] In the context of live-streaming e-commerce, traditional rules or classification models can be used to identify the content of the live stream in order to obtain the product identification results. However, these technologies are difficult to detect hidden misleading information and suggestive information about the products during the live stream. Summary of the Invention

[0003] This disclosure provides a method, apparatus, electronic device, storage medium, and computer program product for identifying goods.

[0004] According to a first aspect of this disclosure, a method for identifying a product is proposed, comprising: acquiring multimodal data during a live broadcast; performing semantic parsing on the multimodal data to obtain structured semantic data; acquiring structured claim data for a live-streamed product based on the structured semantic data; acquiring factual information about the live-streamed product; identifying the live-streamed product based on the structured semantic data, the structured claim data, and the factual information; and obtaining an identification result for the live-streamed product.

[0005] According to a second aspect of this disclosure, a product identification device is proposed, comprising: a first acquisition module, configured to acquire multimodal data during a live broadcast, perform semantic parsing on the multimodal data, and acquire structured semantic data; a second acquisition module, configured to acquire structured claim data for a live-streamed product based on the structured semantic data; a third acquisition module, configured to acquire factual information about the live-streamed product; and an identification module, configured to identify the live-streamed product based on the structured semantic data, the structured claim data, and the factual information, and acquire an identification result for the live-streamed product.

[0006] According to a third aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the product identification method proposed in the first aspect above.

[0007] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to perform the product identification method proposed in the first aspect above.

[0008] According to a fifth aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the product identification method proposed in the first aspect above.

[0009] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0010] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0011] Figure 1 This is a schematic flowchart of a product identification method according to an embodiment of the present disclosure; Figure 2 This is a schematic flowchart of a product identification method according to an embodiment of the present disclosure; Figure 3 This is a schematic diagram of the structure of a product identification device according to an embodiment of the present disclosure; Figure 4 This is a schematic block diagram of an electronic device according to an embodiment of the present disclosure. Detailed Implementation

[0012] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0013] Artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It involves both hardware and software technologies. AI hardware technologies generally include computer vision, speech recognition, natural language processing, and related technologies such as deep learning, big data processing, and knowledge graphs.

[0014] Large models refer to machine learning models with a large number of parameters and high complexity. They require a lot of computing resources and storage space for training and storage, and often require distributed computing and special hardware acceleration technologies. Large models have stronger generalization and expressive capabilities.

[0015] Natural Language Processing (NLP) is an important research direction in the field of artificial intelligence. It integrates knowledge from multiple disciplines such as linguistics, computer science, machine learning, mathematics, and cognitive psychology. It is an interdisciplinary field that combines computer science, artificial intelligence, and linguistics, encompassing two main aspects: natural language understanding and natural language generation. Its research content includes multiple levels such as characters, words, phrases, sentences, paragraphs, and texts, serving as a bridge between machine language and human language. Its aim is to enable machines to understand, interpret, and generate human language, achieving effective communication between humans and machines, and enabling computers to perform tasks such as language translation, sentiment analysis, and text summarization.

[0016] Figure 1 This is a schematic flowchart illustrating a product identification method according to an embodiment of this disclosure. Figure 1 As shown, the method includes: S101: Acquire multimodal data during the live broadcast, perform semantic parsing on the multimodal data, and obtain structured semantic data.

[0017] It should be noted that the executing entity of the product identification method in this embodiment of the disclosure can be a hardware device with the ability to identify data products and / or the necessary software to drive the hardware device to work. Optionally, the executing entity may include a workstation, server, computer, user terminal, and other smart devices. Among them, the user terminal includes, but is not limited to, mobile phones, computers, smart voice interaction devices, smart home appliances, vehicle terminals, etc.

[0018] It should be noted that the multimodal data during a live stream includes at least audio stream data, video stream data, and timestamp information. The video stream data is used to assist in speaker recognition and audio alignment during the live stream.

[0019] Optionally, audio stream data, video stream data, etc. output by the live streaming system can be acquired in real time or offline to obtain multimodal data during the live streaming process.

[0020] In this embodiment of the disclosure, speech signal extraction is performed on multimodal data to obtain the broadcaster's speech signal in the multimodal data. The broadcaster's speech signal is then processed into speech-to-text to obtain a text segment carrying a timestamp. The text segment is then semantically segmented to obtain at least one corresponding semantic unit and key information of the semantic unit, thereby obtaining structured semantic data.

[0021] S102, Based on the structured semantic data, obtain the structured claim data for the live-streamed products during the live-streaming process.

[0022] It should be noted that after obtaining structured semantic data, the structured semantic data can be identified to determine the semantic claims on the live-streamed products, thereby obtaining structured claim data for the live-streamed products during the live-streaming process.

[0023] It should be noted that the identification of structured semantic data is not a traditional and simple text classification, but rather an ability to identify "word games" and "suggestive information" during live streaming.

[0024] Optionally, based on a large model, structured semantic data can be identified to obtain explicit claims, implicit claims, or suggestive conclusions of the anchor regarding the product, thereby obtaining the claim data of the live-streamed product. The claim data can then be stored in a structured manner to obtain the structured claim data of the live-streamed product.

[0025] In this context, explicit claim data for live-streamed products can be understood as product data explicitly expressed by the live-streamer through direct and clear language; implicit claim data for live-streamed products can be understood as product data indirectly expressed by the live-streamer through hints, metaphors, or contextual associations.

[0026] S103, Obtain factual information about the products sold in the live stream.

[0027] In this context, the factual information about products sold in live streams can be understood as the objective attribute information of the products.

[0028] Optionally, factual data sources of the live-streamed products can be obtained, and information can be extracted from the factual data sources to obtain factual information about the live-streamed products.

[0029] It should be noted that the factual data sources for live-streamed products include, but are not limited to, the product's instruction manual, its registration information, its tags and category attributes, and the brand's official information.

[0030] It should be noted that the information extracted includes, but is not limited to, the efficacy claims, ingredient information, target audience, usage conditions, contraindications, and regulatory restrictions of the products featured in the live stream.

[0031] For example, an information extraction model can be used to extract information from the factual data source of the live-streamed product based on the information extraction object, thereby obtaining the factual triples of the live-streamed product and thus acquiring the factual information of the live-streamed product.

[0032] For example, regarding the instruction manual text of product A in a live stream, which states "This product contains hyaluronic acid, is suitable for dry skin, and should be used with caution by pregnant women," an information extraction model is used to extract information from the instruction manual text of product A based on the information extraction target. This yields the factual triplet for product A: {Product A, contains ingredients, hyaluronic acid}, {Product A, suitable for, dry skin}, and {Product A, contraindicated for, pregnant women}. The above process is repeated for other factual data sources related to product A in the live stream to obtain factual information about the product.

[0033] S104, Identify live-streamed products based on structured semantic data, structured claim data, and factual information, and obtain the identification results of live-streamed products.

[0034] In this embodiment of the disclosure, structured claim data and factual information can be aligned to obtain the alignment result of claims and facts for live-streamed products. Based on the structured semantic data, factual information, and alignment result, live-streamed products can be identified to obtain the risk identification result of live-streamed products.

[0035] In this embodiment of the disclosure, a large model is used to simulate the user's understanding process based on the claims and contextual dependency information in the structured semantic data, so as to generate a set of potential conclusions of the user regarding the live-streamed products. The set of potential conclusions is then checked for erroneous cognition based on factual information to obtain the check result of the set of potential conclusions. Based on the check result and the alignment result, the recognition result of the live-streamed products is obtained.

[0036] The product identification method proposed in this disclosure acquires multimodal data during a live stream, performs semantic parsing on the multimodal data to obtain structured semantic data, and based on the structured semantic data, acquires structured claim data regarding the live stream products, as well as factual information about the live stream products. The live stream products are then identified based on the structured semantic data, structured claim data, and factual information to obtain the identification results. Therefore, this disclosure, by identifying live stream products based on structured semantic data, structured claim data, and factual information, can accurately identify implicit misleading information present in live stream products, improving the accuracy of live stream product identification and contributing to enhancing the authenticity of live stream product information.

[0037] Figure 2 This is a schematic flowchart of a product identification method according to a second embodiment of the present disclosure.

[0038] like Figure 2 As shown, in Figure 2 Based on the illustrated embodiments, the product identification method of this disclosure may specifically include the following steps: S201, acquire multimodal data during the live broadcast.

[0039] Optionally, step S101 "perform semantic parsing on multimodal data to obtain structured semantic data" in the above embodiments may specifically include the following steps S202-S204.

[0040] S202, extract speech signals from multimodal data to obtain the broadcaster's speech signal from the multimodal data.

[0041] It should be noted that in live streaming scenarios, given the presence of background music, video sound effects, and other audio signals, to ensure the accuracy of the acquired structured semantic data, a speaker separation model can be used to extract audio signals from multimodal data, distinguishing between the broadcaster's voice and non-broadcaster's voice, and representing the audio signal as a time function. Extract the set of audio segments that match the broadcaster, using the broadcaster's voice signal.

[0042] S203 performs speech-to-text processing on the broadcaster's voice signal to obtain a text segment with a timestamp.

[0043] Optionally, a speech recognition model can be used to process the broadcaster's voice signal into text, obtaining text segments with timestamps: ,in, For the j-th word or phrase, For the start timestamp corresponding to the j-th word or phrase, Let be the end timestamp corresponding to the j-th word or phrase.

[0044] It should be noted that, in order to ensure the quality of the text fragments, invalid information can be removed. By constructing a set of noisy words for the live streaming scene, text cleaning and noise filtering can be performed on the text fragments based on the set of noisy words for the live streaming scene, in order to remove content in the text fragments that does not actually contribute to semantic understanding, such as: colloquialisms, repetitions, and small talk.

[0045] For example, a set of noisy words for live streaming scenarios. Based on the set of noisy words in a live streaming scenario, the process of text cleaning and noise filtering for text fragments can be represented as follows: .

[0046] For example, the original text fragment is "Come on, brothers, this product, this, really a lot of people are using it." After cleaning and noise filtering the original text fragment, we get "This product is used by many people." S204, perform semantic segmentation on the text fragment, obtain at least one corresponding semantic unit and key information of the semantic unit, so as to obtain structured semantic data.

[0047] It should be noted that live streaming scripts are characterized by long sentences, multiple semantic components, and strong contextual dependencies. Therefore, directly segmenting text fragments according to periods will affect the subsequent product recognition results.

[0048] In this embodiment of the disclosure, a segmentation method based on semantic intent change can be used to obtain multiple text segments. The semantic similarity between adjacent text segments is determined according to the semantic vector similarity function. When the semantic similarity is less than a preset similarity threshold, the text segments are semantically segmented to obtain at least one corresponding semantic unit.

[0049] For example, if Then determine and They belong to different semantic units, among which, For the i-th text segment, For the (i+1)th text segment, () is a semantic similarity function, This is the similarity threshold.

[0050] For example, for the text fragment "I use this face mask every day, and it's suitable for sensitive skin. Many people have reported good results," semantic segmentation of the text fragment yields three semantic units: semantic unit 1 "I use this face mask every day," semantic unit 2 "It's suitable for sensitive skin," and semantic unit 3 "The feedback on its effectiveness is good."

[0051] In this embodiment of the disclosure, at least one semantic unit corresponding to a text fragment and a timestamp corresponding to the semantic unit are obtained, a context window centered on the semantic unit is determined, and a context window index is assigned to the semantic unit, and a product identifier of the semantic unit is obtained; wherein, the key information of the semantic unit includes semantic text content, timestamp, context window index and product identifier.

[0052] Optionally, the size of the sliding window can be obtained, and the context window centered on the semantic unit can be determined based on the size of the sliding window.

[0053] For example, if the size of the sliding window is m, for the k-th semantic unit... The context window centered on the k-th semantic unit is ,in, For the context window of the k-th semantic unit, For the first A semantic unit, For the first A semantic unit, This is the size of the sliding window.

[0054] In this embodiment of the disclosure, after obtaining the semantic text content, timestamp, context window index, and product identifier, structured semantic data can be constructed: ,in, The i-th semantic unit is a set of key information for a semantic unit. ,in, For the first The semantic text content of each semantic unit For the first The start timestamp of each semantic unit For the first End timestamp of each semantic unit For the first Context window index of each semantic unit, For the first Product identifiers for semantic units.

[0055] Optionally, step S102 in the above embodiments, "obtaining structured claim data for live-streamed products based on structured semantic data", may specifically include the following steps S205-S206.

[0056] S205 identifies structured semantic data and generates candidate claim information for live-streamed products.

[0057] In this embodiment of the disclosure, structured semantic data can be identified based on a large model and preset prompts to generate candidate claim information for live-streamed products.

[0058] S206, Generate structured claim data based on candidate claim information.

[0059] In this embodiment of the disclosure, candidate claim information is identified, claim object and claim type are obtained, claim content of claim information is obtained, implicit or explicit claim is identified in the claim content of candidate claim information, claim tag is obtained, claim strength of candidate claim information is determined, and context dependency information of candidate claim information is obtained.

[0060] The structured claim data includes at least the claim object, claim type, claim content, claim tag, claim strength, and contextual dependency information of the candidate claim information.

[0061] In this embodiment of the disclosure, structured claim data can be represented as:

[0062] in, For structured claim data, Structured claim data of the j-th semantic unit , For the claim content of the j-th semantic unit, For the claim type of the j-th semantic unit, For the claim object of the j-th semantic unit, The claim strength of the j-th semantic unit This represents the context dependency information for the j-th semantic unit.

[0063] In this embodiment of the disclosure, entity recognition is performed on the candidate claim information to obtain the claim object, and type recognition is performed on the candidate claim information based on a predefined claim type space to obtain the claim type.

[0064] Optionally, a predefined claim type space T{efficacy claims, safety claims, applicability claims, comparative claims, endorsement claims} is obtained, and the probability of a semantic unit belonging to each claim type is determined by the large model based on the structured semantic data of the semantic unit.

[0065] For example, a large model can determine the probability that a semantic unit belongs to each claim type based on the structured semantic data of the semantic unit and the claim type space T. And select the claim type with the highest probability. ,Right now

[0066] In this embodiment of the disclosure, candidate claim information is analyzed to determine a first parameter of the candidate claim information. The first parameter includes at least the degree of certainty of expression, the frequency of repetition, and the degree of emotionality. The claim strength of the claim information is determined based on the parameters in the first parameter and the corresponding weight values ​​of the parameters.

[0067] Among them, the different strengths of the assertions in the information have varying degrees of psychological impact on users.

[0068] Optionally, a claim strength calculation model can be constructed, and the claim strength of the claim information can be determined based on the parameters and their corresponding weight values ​​in the first parameter. The expression for the claim strength calculation model is as follows:

[0069] in, To advocate strength, To express the degree of certainty, To express the weight values ​​corresponding to the degree of certainty, For recurrence frequency, The weight value corresponding to the frequency of recurrence For the degree of emotionality, This represents the weight value corresponding to the degree of emotionality.

[0070] It's worth noting that in live-streaming scenarios, when hosts introduce products, they often don't state the claims directly, but rather imply and guide users to make their own judgments. This can be achieved by building a claim tag prediction model to determine the content of the claim. Does it belong to The criteria for judgment include, but are not limited to, the presence of evaluative verbs, the presence of omitted conditions, and the presence of emotional implications. If the content of the claim... belong If the claim label is 0, it can be understood as an explicit claim; if the claim content... Not belonging to If the label is 1, it can be understood as an implicit claim.

[0071] The expression for the proposed label prediction model is as follows:

[0072] in, For the structured semantic data of the j-th semantic unit, For the claim content of the j-th semantic unit, The claim label for the j-th semantic unit.

[0073] For example, regarding semantic unit 1 "I use this face mask every day," it is suggestive and therefore judged as an implicit claim. .

[0074] It should be noted that a single semantic statement is often insufficient to obtain the identification result of a product. It is necessary to combine contextual information. In this embodiment of the disclosure, a context window centered on the target semantic unit is determined based on the context window index. The target semantic unit is a semantic unit that contains candidate claim information. Within the context window, the context semantic units associated with the target semantic unit are identified, and the context claim information is extracted from the context semantic units as the context dependency information of the candidate claim information.

[0075] For example, other claims preceding candidate claims can be used as precondition claims, and other claims following candidate claims can be used as subsequent supplementary claims. By examining all candidate claims within the context window, it can be determined whether any key conditions are missing, and these are used as key condition claims to obtain contextual dependency information for the candidate claims.

[0076] S207, Obtain factual information about products sold in the live stream.

[0077] Optionally, step S104 in the above embodiments, "identifying live-streamed products based on structured semantic data, structured claim data, and factual information, and obtaining the identification results of live-streamed products," may specifically include the following steps S208-S209.

[0078] S208 aligns structured claim data with factual information to obtain the alignment results of claims and facts for live-streamed products.

[0079] In this embodiment of the disclosure, the fact dimension corresponding to the fact information in the standard fact dimension space is obtained, and structured fact information is constructed based on the fact dimension, fact value and constraints of the fact information. The structured claim data is aligned with the structured fact information to obtain the alignment result of the product's claim and facts.

[0080] It should be noted that by uniformly mapping factual information to a standard factual dimension space, it is easier to obtain the alignment results between the product's claims and the facts.

[0081] The standard fact dimension space D includes efficacy, ingredients, safety, suitability, conditions of use, and regulatory restrictions.

[0082] In this embodiment of the disclosure, structured fact information can be constructed based on the fact dimensions, fact values, and constraints of the fact information. The structured fact information can be represented as follows: ,in, For the fact dimension of the k-th fact information, The fact value of the kth fact information The constraint condition for the k-th factual information is... This is the k-th structured fact information.

[0083] Optionally, based on the semantic mapping function, the fact dimension corresponding to the structured claim data in the standard fact dimension space can be obtained. Under the same fact dimension, the claim content of the structured claim data is aligned with the structured fact information to obtain the alignment status of the product's claim and facts. The alignment basis information of the alignment status is obtained. Based on the alignment basis information, the cause information and evidence information associated with the alignment status are determined. The cause information and evidence information are associated with the alignment status to generate the alignment result.

[0084] For example, it can be based on semantic mapping functions This involves mapping structured claim data to a standard fact dimension space to determine the most relevant fact dimension for the structured data within that space. , which is the fact dimension corresponding to structured claim data in the standard fact dimension space.

[0085] In this embodiment of the disclosure, an alignment state is determined in response to structured fact information being supported by fact information. For complete alignment; in response to the existence of constraints in the structured factual information and the absence of constraints in the claim content, determine the alignment state. For partial alignment; in response to contradictions between structured factual information and claims, determine the alignment status. To resolve alignment conflicts.

[0086] It should be noted that by associating causal information and evidentiary information with the alignment status to generate alignment results, traceable evidence can be provided for the identification results of subsequent live-streamed products.

[0087] S209. Based on structured semantic data, factual information, and alignment results, identify the live-streamed products and obtain the identification results of the live-streamed products.

[0088] In this embodiment of the disclosure, a large model is used to simulate the user's understanding process based on the claims and contextual dependency information in the structured semantic data, so as to generate a set of potential conclusions about the product. The set of potential conclusions is then checked for errors based on factual information to obtain the verification result of the set of potential conclusions. Based on the verification result and the alignment result, the product identification result is obtained.

[0089] It should be noted that when the host does not directly state the false information about the product, users may automatically infer their understanding of the product based on the context. A large model can be used to simulate the user's understanding process based on the claims and contextual dependency information in the structured semantic data, in order to generate a set of potential conclusions about the product from the user.

[0090] For example, simulating a user's understanding process can be represented as: ,in, For the set of potential conclusions, For the content of the claim, This is context-dependent information.

[0091] In this embodiment of the disclosure, a set of supportable conclusions corresponding to factual information is constructed based on factual information. Based on the set of supportable conclusions, each potential conclusion in the potential conclusion set is checked for erroneous cognition to obtain the check result of the potential conclusion set.

[0092] The verification results of the potential conclusion set can be obtained in the following ways:

[0093] in, The set of supporting conclusions corresponding to factual information. These are potential conclusions within the set of potential conclusions.

[0094] It should be noted that if the potential conclusions in the potential conclusion set belong to To determine if the potential conclusions are not misleading, if the potential conclusions in the set of potential conclusions do not belong to... This suggests that potential conclusions may be misleading.

[0095] In this embodiment of the disclosure, in response to the presence of potential conclusions in the verification results that do not belong to the set of supported conclusions, the risk type of the live-streaming product is determined from a preset risk type set based on the alignment results and structured claim data, wherein the identification result of the live-streaming product also includes the risk type of the live-streaming product.

[0096] It should be noted that the set of risk types includes, but is not limited to, risks of incomplete information, risks of misrepresentation, risks of implied promises, and risks of selective disclosure.

[0097] In this embodiment of the disclosure, the risk type of a live-streamed product can be determined from a preset set of risk types based on the alignment results and the claim tags and contextual dependency information in the structured claim data.

[0098] in, ,in, For the alignment status in the alignment result, To advocate for labels, This is context-dependent information.

[0099] In this embodiment of the disclosure, a second parameter of the live-streaming product is determined based on the claim label, claim strength, and context dependency information. The second parameter includes at least the degree of claim implicature, claim strength, and context coverage. The risk score of the live-streaming product is determined based on the parameters and their corresponding weight values. The identification result of the live-streaming product also includes the risk score of the live-streaming product.

[0100] Optionally, a risk score calculation model can be constructed, and the risk score of the live-streamed product can be determined based on the parameters and their corresponding weights in the second parameter. The expression for calculating the risk score is as follows:

[0101] in, For risk scoring, To assert the degree of implied meaning, To assert the weight corresponding to the degree of implicitness, Advocating strength, The weight corresponding to the strength of the claim, For context coverage, The weights corresponding to the degree of context coverage.

[0102] In this embodiment of the disclosure, the risk cause of a product can be determined based on the alignment result. The identification result of the live-streamed product also includes the risk cause of the live-streamed product. .

[0103] In this embodiment of the disclosure, the identification result of the live-streamed product can be represented as:

[0104] In this embodiment of the disclosure, the risk level of the live-streamed product is determined based on the risk score, the risk cause is decomposed in a structured manner to obtain the structured risk cause field, and a risk warning text is generated based on the predefined prompt template, the structured risk cause field, the risk level, and the risk type. Based on the alignment result, the evidence information corresponding to the risk warning text is obtained, and the risk warning information of the live-streamed product is generated based on the risk warning text and the evidence information of the risk warning text.

[0105] Optionally, a preset risk threshold can be obtained, and the risk level of the live-streamed product can be determined based on the risk threshold, making it easier for users to understand and make decisions.

[0106] For example, setting the first risk threshold to... Second risk threshold ,when At that time, determine the risk level. For low risk, when At that time, determine the risk level. For medium risk, when At that time, determine the risk level. It is considered high-risk.

[0107] It should be noted that, in order to avoid generating subjective prompts that are untraceable and unverifiable, the risk causes can be decomposed in a structured manner to obtain the structured risk cause field.

[0108] It should be noted that the reasons for the risks... Decomposed into structured fields, resulting in structured risk reason fields: ,in, For the type of risk cause, For missing or conflicting factual elements, This creates a path of misunderstanding for the user.

[0109] It should be noted that directly using large models to freely generate risk warning information often introduces new misleading information. To avoid "confused or misleading statements" and ensure the objectivity and standardization of risk warning texts, risk warning texts can be generated based on predefined warning templates, structured risk cause fields, risk levels, and risk types. For example, the prompt template could be: "The host did not explain [missing_fact] when introducing this product," "This may lead consumers to misunderstand [misleading_path]," or "It is recommended to pay attention to relevant restrictions before purchasing." It should be noted that, in order to improve the credibility of the identification results for live-streamed products and support subsequent traceability, evidence information corresponding to the risk warning text can be obtained based on the alignment results. (e.g., the original text of the product facts cited, the corresponding claim text, the start timestamp and the end timestamp).

[0110] In this embodiment of the disclosure, the target output object for obtaining risk warning information is selected, the risk warning information is adjusted according to the output object to obtain target risk warning information, and the target risk warning information is pushed to the target output object.

[0111] It should be noted that by pushing the target risk warning information to the target output object, the understandability and usability of the product recognition results can be improved, achieving accurate adaptation of the target risk warning information to the target output object. It should be noted that the same risk warning information should be expressed differently to different recipients, including but not limited to users and live streaming platforms.

[0112] For example, if the target audience is users, the target risk warning message is "The host did not explain that this effect only works under specific conditions, please judge with caution"; if the target audience is the live streaming platform, the target risk warning message is "Incomplete disclosure of efficacy information was detected, it is recommended to review the live streaming content".

[0113] In summary, the product identification method proposed in this disclosure involves acquiring multimodal data during a live stream, extracting speech signals from the multimodal data, obtaining the anchor's speech signal from the multimodal data, performing speech-to-text processing on the anchor's speech signal to obtain text segments with timestamps, performing semantic segmentation on the text segments to obtain at least one corresponding semantic unit and key information of the semantic unit to obtain structured semantic data, identifying the structured semantic data to generate candidate claim information for the live stream product, generating structured claim data based on the candidate claim information, obtaining factual information about the live stream product, aligning the structured claim data with the factual information, and obtaining the alignment result of the claim and fact of the live stream product. This invention simulates the user's understanding process based on a large model, using the claims and contextual dependencies in structured semantic data to generate a set of potential conclusions about a product. It then verifies these potential conclusions against factual information to obtain a verification result. Based on the verification and alignment results, it obtains the product's identification result. Therefore, by simulating the user's understanding process to generate a set of potential conclusions about a product, this invention can accurately identify implicit misleading information, incomplete information, and suggestive promises in products. This improves the accuracy, interpretability, and reliability of identifying the true information of live-streamed products and can be used in applications such as real-time monitoring of live-streaming platforms, consumer risk warnings, and regulatory compliance assistance.

[0114] It should be noted that the information (including but not limited to Q&A information, user device information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.) and signals involved in this disclosure are all authorized by the user or fully authorized by all parties, and the source, use and processing of the relevant data comply with relevant laws, regulations and standards, and do not violate public order and good morals.

[0115] According to embodiments of this disclosure, this disclosure also provides a product identification device for implementing the above-described product identification method.

[0116] Figure 3 This is a block diagram of a product identification device according to an embodiment of the present disclosure.

[0117] like Figure 3 As shown, the product identification device 300 includes: a first acquisition module 301, a second acquisition module 302, a third acquisition module 303, and an identification module 304.

[0118] The first acquisition module 301 is used to acquire multimodal data during the live broadcast, perform semantic parsing on the multimodal data, and acquire structured semantic data. The second acquisition module 302 is used to acquire structured claim data for live-streamed products during the live-streaming process based on the structured semantic data. The third acquisition module 303 is used to acquire factual information about the live-streamed products; The identification module 304 is used to identify the live-streamed product based on the structured semantic data, the structured claim data, and the factual information, and to obtain the identification result of the live-streamed product.

[0119] In one embodiment of this disclosure, the identification module 304 is configured to: align the structured claim data with the factual information to obtain the alignment result of the claim and fact of the live-streamed product; and identify the live-streamed product based on the structured semantic data, the factual information, and the alignment result to obtain the identification result of the live-streamed product.

[0120] In one embodiment of this disclosure, the identification module 304 is configured to: obtain the fact dimension corresponding to the fact information in a standard fact dimension space; construct structured fact information based on the fact dimension, fact value, and constraints of the fact information; align the structured claim data with the structured fact information to obtain the alignment result of the claim and facts of the live-streamed product.

[0121] In one embodiment of this disclosure, the identification module 304 is configured to: obtain the fact dimension corresponding to the structured claim data in the standard fact dimension space according to the semantic mapping function; align the claim content of the structured claim data with the structured fact information under the same fact dimension to obtain the alignment status of the claim and fact of the live-streamed product; obtain alignment basis information of the alignment status; determine the cause information and evidence information associated with the alignment status according to the alignment basis information; associate the cause information and the evidence information with the alignment status to generate the alignment result.

[0122] In one embodiment of this disclosure, the identification module 304 is configured to: determine that the alignment state is fully aligned in response to the structured fact information being supported by the fact information; or determine that the alignment state is partially aligned in response to the structured fact information having a limiting condition and the claim content not having the limiting condition; or determine that the alignment state is an alignment conflict in response to the structured fact information contradicting the claim content.

[0123] In one embodiment of this disclosure, the first acquisition module 301 is configured to: extract speech signals from the multimodal data to acquire the broadcaster's speech signals in the multimodal data; perform speech-to-text processing on the broadcaster's speech signals to acquire text segments carrying timestamps; and perform semantic segmentation on the text segments to acquire at least one corresponding semantic unit and key information of the semantic unit, thereby acquiring the structured semantic data.

[0124] In one embodiment of this disclosure, the first acquisition module 301 is configured to: acquire at least one semantic unit corresponding to the text fragment and a timestamp corresponding to the semantic unit; determine a context window centered on the semantic unit and assign a context window index to the semantic unit; acquire the product identifier of the semantic unit; wherein, the key information of the semantic unit includes semantic text content, timestamp, context window index and product identifier.

[0125] In one embodiment of this disclosure, the second acquisition module 302 is configured to: identify the structured semantic data and generate candidate claim information for the live-streamed product; and generate the structured claim data based on the candidate claim information.

[0126] In one embodiment of this disclosure, the second acquisition module 302 is configured to: identify the candidate claim information to obtain the claim object and claim type; obtain the claim content of the claim information, identify implicit or explicit claims in the claim content of the candidate claim information to obtain claim tags; determine the claim strength of the candidate claim information; and obtain the context dependency information of the candidate claim information; the structured claim data includes at least the claim object, claim type, claim content, claim tags, claim strength, and context dependency information of the candidate claim information.

[0127] In one embodiment of this disclosure, the second acquisition module 302 is configured to: perform entity recognition on the candidate claim information to obtain the claim object; and perform type recognition on the candidate claim information based on a predefined claim type space to obtain the claim type.

[0128] In one embodiment of this disclosure, the second acquisition module 302 is configured to: analyze the candidate claim information, determine a first parameter of the candidate claim information, wherein the first parameter includes at least the degree of certainty of expression, the frequency of repetition, and the degree of emotionality; and determine the claim strength of the claim information based on the parameters in the first parameter and the weight values ​​corresponding to the parameters.

[0129] In one embodiment of this disclosure, the second acquisition module 302 is configured to: determine a context window centered on a target semantic unit based on the context window index, wherein the target semantic unit is a semantic unit containing candidate claim information; identify context semantic units associated with the target semantic unit within the context window; and extract context claim information from the context semantic units as context dependency information of the candidate claim information.

[0130] In one embodiment of this disclosure, the identification module 304 is configured to: simulate the user's understanding process based on a large model according to the claim content and contextual dependency information in the structured semantic data, so as to generate a set of potential conclusions of the user regarding the live-streamed product; perform error cognition verification on the set of potential conclusions based on the factual information to obtain a verification result of the set of potential conclusions; and obtain the identification result of the live-streamed product based on the verification result.

[0131] In one embodiment of this disclosure, the identification module 304 is configured to: construct a set of supportable conclusions corresponding to the factual information based on the factual information; and perform error cognition verification on each potential conclusion in the potential conclusion set based on the set of supportable conclusions to obtain the verification result of the potential conclusion set.

[0132] In one embodiment of this disclosure, the identification module 304 is configured to: in response to the presence of a potential conclusion in the verification result that does not belong to the set of supported conclusions, determine the risk type of the live-streaming product from a preset risk type set based on the alignment result and the structured claim data, wherein the identification result of the live-streaming product also includes the risk type of the live-streaming product.

[0133] In one embodiment of this disclosure, the apparatus 300 is configured to: determine a second parameter of the live-streamed product based on the claim tag, the claim strength, and the context dependency information, wherein the second parameter includes at least the claim implicature, claim strength, and context coverage; determine a risk score of the live-streamed product based on the parameters in the second parameter and the weight values ​​corresponding to the parameters, wherein the identification result of the live-streamed product also includes the risk score of the live-streamed product.

[0134] In one embodiment of this disclosure, the apparatus 300 is configured to: determine the risk cause of the live-streamed product based on the alignment result; the identification result of the live-streamed product further includes the risk cause of the live-streamed product.

[0135] In one embodiment of this disclosure, after obtaining the identification result of the live-streamed product, the device 300 is configured to: determine the risk level of the live-streamed product based on the risk score; perform structured decomposition of the risk cause to obtain a structured risk cause field; generate a risk warning text based on a predefined prompt template, the structured risk cause field, the risk level, and the risk score; obtain evidence information corresponding to the risk warning text based on the alignment result; and generate risk warning information for the live-streamed product based on the risk warning text and the evidence information of the risk warning text.

[0136] In one embodiment of this disclosure, the apparatus 300 is configured to: obtain a target output object of the risk warning information; adjust the risk warning information according to the output object to obtain target risk warning information; and push the target risk warning information to the target output object.

[0137] The product identification device disclosed herein acquires multimodal data during a live broadcast, performs semantic parsing on the multimodal data to obtain structured semantic data, and based on the structured semantic data, acquires structured claim data regarding the live-streamed product, as well as factual information about the live-streamed product. It then identifies the live-streamed product based on the structured semantic data, structured claim data, and factual information, obtaining the identification result. Therefore, this disclosure, by identifying live-streamed products based on structured semantic data, structured claim data, and factual information, can accurately identify implicit misleading information present in live-streamed products, improving the accuracy of product identification and enhancing the authenticity of live-streamed product information.

[0138] According to embodiments of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0139] Figure 4 A schematic block diagram of an example electronic device 400 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0140] like Figure 4 As shown, device 400 includes a computing unit 401, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 402 or a computer program loaded from storage unit 408 into random access memory (RAM) 403. RAM 403 may also store various programs and data required for the operation of device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0141] Multiple components in device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of monitors, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0142] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as a product identification method. For example, in some embodiments, the product identification method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of the product identification method described above may be performed. Alternatively, in other embodiments, the computing unit 401 may be configured to perform the product identification method by any other suitable means (e.g., by means of firmware).

[0143] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0144] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0145] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0146] To initiate interaction with a user account, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user account; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user account can submit input to the computer. Other types of devices can also be used to initiate interaction with the user account; for example, feedback submitted to the user account can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user account can be received in any form (including voice input, speech input, or tactile input).

[0147] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user account computer with a graphical user interface or web browser through which a user account can interact with the implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0148] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0149] According to embodiments of this disclosure, this disclosure also provides a computer program product, including a computer program, wherein when the computer program is executed by a processor, it implements the steps of the commodity identification method described in the above embodiments of this disclosure.

[0150] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0151] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for identifying a commodity, characterized in that, The method includes: Acquire multimodal data during the live broadcast, perform semantic parsing on the multimodal data, and obtain structured semantic data; Based on the structured semantic data, obtain structured claim data for live-streamed products during the live-streaming process; Obtain factual information about the products being live-streamed; The live-streamed products are identified based on the structured semantic data, the structured claim data, and the factual information, and the identification results of the live-streamed products are obtained.

2. The method according to claim 1, wherein, The step of identifying the live-streamed product based on the structured semantic data, the structured claim data, and the factual information, and obtaining the identification result of the live-streamed product, includes: Align the structured claim data with the factual information to obtain the alignment result of the claims and facts of the live-streamed product; Based on the structured semantic data, the factual information, and the alignment results, the live-streamed products are identified, and the identification results of the live-streamed products are obtained.

3. The method according to claim 2, wherein, Aligning the structured claim data with the factual information to obtain the alignment result of the claims and facts for the live-streamed product includes: Obtain the fact dimension corresponding to the fact information in the standard fact dimension space; Based on the fact dimensions, fact values, and constraints of the aforementioned fact information, structured fact information is constructed; Align the structured claim data with the structured fact information to obtain the alignment result of the claims and facts of the live-streamed product.

4. The method according to claim 3, wherein, The process of aligning the structured claim data with the structured fact information to obtain the alignment result of the claims and facts of the live-streamed product includes: Based on the semantic mapping function, obtain the fact dimension corresponding to the structured claim data in the standard fact dimension space; Under the same fact dimension, the claim content of the structured claim data is aligned with the structured fact information to obtain the alignment status of the claim and fact of the live-streamed product; Obtain the alignment basis information for the alignment state; Based on the alignment criteria information, determine the cause information and evidence information associated with the alignment state; The alignment result is generated by associating the cause information and the evidence information with the alignment state.

5. The method according to claim 4, wherein, Aligning the claim content of the structured claim data with the structured fact information to obtain the alignment status of the claims and facts of the live-streamed product includes: In response to the structured fact information being supported by the fact information, the alignment state is determined to be fully aligned; or In response to the existence of constraints in the structured fact information, and the absence of such constraints in the claim content, the alignment state is determined to be partial alignment; or In response to a contradiction between the structured factual information and the claim, the alignment state is determined to be an alignment conflict.

6. The method according to claim 1, wherein, The step of performing semantic parsing on the multimodal data to obtain structured semantic data includes: Speech signal extraction is performed on the multimodal data to obtain the broadcaster's speech signal from the multimodal data; The broadcaster's voice signal is processed to convert speech to text, and a text segment carrying a timestamp is obtained. The text segment is semantically segmented to obtain at least one corresponding semantic unit and key information of the semantic unit, thereby obtaining the structured semantic data.

7. The method according to claim 6, wherein, The step of semantically segmenting the text fragment to obtain at least one corresponding semantic unit and key information of the semantic unit includes: Obtain at least one semantic unit corresponding to the text fragment and the timestamp corresponding to the semantic unit; Determine the context window centered on the semantic unit and assign a context window index to the semantic unit; Obtain the product identifier of the semantic unit; The key information of the semantic unit includes semantic text content, timestamp, context window index, and product identifier.

8. The method according to claim 1, wherein, The step of obtaining structured claim data for live-streamed products based on the structured semantic data includes: The structured semantic data is identified to generate candidate claim information for the live-streamed products; The structured claim data is generated based on the candidate claim information.

9. The method according to claim 8, wherein, The step of generating the structured claim data based on the candidate claim information includes: The candidate claim information is identified to obtain the claim object and claim type; Obtain the content of the claims in the claim information, identify implicit or explicit claims in the content of the claims in the candidate claim information, and obtain claim tags; Determine the claim strength of the candidate claim information; Obtain the contextual dependency information of the candidate claim information; The structured claim data includes at least the claim object, claim type, claim content, claim tag, claim strength, and context dependency information of the candidate claim information.

10. The method according to claim 9, wherein, The step of identifying the candidate claim information and obtaining the claim type and claim object includes: Entity identification is performed on the candidate claim information to obtain the claim object; The candidate claim information is type-identified based on a predefined claim type space to obtain the claim type.

11. The method according to claim 9, wherein, Determining the claim strength of the candidate claim information includes: The candidate claim information is analyzed to determine a first parameter of the candidate claim information, wherein the first parameter includes at least the degree of certainty of expression, the frequency of repetition, and the degree of emotionality; The claim strength of the claim information is determined based on the parameters in the first parameter and the weight values ​​corresponding to the parameters.

12. The method according to claim 9, wherein, The contextual dependency information for obtaining the candidate claim information includes: Based on the context window index, a context window centered on the target semantic unit is determined, wherein the target semantic unit is a semantic unit containing candidate claim information; Within the context window, identify the context semantic unit associated with the target semantic unit; Contextual claim information is extracted from the contextual semantic unit and used as contextual dependency information for the candidate claim information.

13. The method according to any one of claims 1-12, wherein, The step of identifying the live-streamed product based on the structured semantic data, the factual information, and the alignment result, and obtaining the identification result of the live-streamed product, includes: Based on the large model, the user's understanding process is simulated according to the claims and contextual dependency information in the structured semantic data, so as to generate a set of potential conclusions of the user regarding the live-streamed products; Based on the factual information, the set of potential conclusions is subjected to error perception verification to obtain the verification result of the set of potential conclusions; Based on the verification result and the alignment result, the recognition result of the live-streamed product is obtained.

14. The method according to claim 13, wherein, The step of performing error perception verification on the potential conclusion set based on the factual information to obtain the verification result of the potential conclusion set includes: Based on the factual information, construct a set of supported conclusions corresponding to the factual information; Based on the set of supported conclusions, each potential conclusion in the set of potential conclusions is subjected to error perception verification to obtain the verification result of the set of potential conclusions.

15. The method according to claim 14, wherein, The method further includes: In response to the presence of potential conclusions in the verification results that do not belong to the set of supported conclusions, the risk type of the live-streaming product is determined from a preset set of risk types based on the alignment results and the structured claim data, wherein the identification result of the live-streaming product also includes the risk type of the live-streaming product.

16. The method according to claim 13, wherein, The method further includes: The second parameter of the live-streamed product is determined based on the claim tag, the claim strength, and the context dependency information, wherein the second parameter includes at least the claim implicature, claim strength, and context coverage. The risk score of the live-streamed product is determined based on the parameters in the second parameter and the corresponding weight values ​​of the parameters. The identification result of the live-streamed product also includes the risk score of the live-streamed product.

17. The method of claim 14, wherein, The method further includes: Based on the alignment results, determine the causes of risk for the live-streamed products; The identification results of the live-streamed products also include the reasons for the risks associated with those products.

18. The method according to claim 13, wherein, After obtaining the identification result of the live-streamed product, the process further includes: The risk level of the live-streamed product is determined based on the risk score. The risk causes are decomposed into a structured form to obtain the structured risk cause field; Based on the predefined prompt template, the structured risk reason field, risk level, and risk score, generate risk prompt text; Based on the alignment result, obtain the evidence information corresponding to the risk warning text; Based on the risk warning text and the evidence information provided by the risk warning text, risk warning information for the live-streamed product is generated.

19. The method according to claim 18, wherein, The method further includes: The target output object for obtaining the risk warning information; Based on the output object, the risk warning information is adjusted to obtain the target risk warning information; The target risk warning information is pushed to the target output object.

20. A product identification device, wherein, The device includes: The first acquisition module is used to acquire multimodal data during the live broadcast, perform semantic parsing on the multimodal data, and acquire structured semantic data. The second acquisition module is used to acquire structured claim data for live-streamed products during the live-streaming process based on the structured semantic data. The third acquisition module is used to acquire the live stream. Factual information about the product; The identification module is used to identify the live-streamed product based on the structured semantic data, the structured claim data, and the factual information, and to obtain the identification result of the live-streamed product.

21. An electronic device, characterized in that, Including processor and memory; The processor runs a program corresponding to the executable program code stored in the memory to implement the method as described in any one of claims 1-19.

22. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-19.

23. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-19.

Images