Object attribute identification method and device, equipment and storage medium

Abstract

The invention discloses an object attribute recognition method and device, equipment and a storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: carrying out the interaction keyword extraction of a plurality of interaction record texts corresponding to a to-be-recognized object, and obtaining a plurality of interaction keywords, each interaction record text is an associated text generated in the process of carrying out resource interaction between the interaction object and the to-be-identified object; inputting the plurality of interaction keywords into a keyword feature extraction model for keyword feature extraction to obtain interaction feature information; and on the basis of the identification text feature information and the interaction feature information of the to-be-identified object, performing merchant attribute identification on the to-be-identified object to obtain merchant attribute indication information used for indicating whether the to-be-identified object has merchant attributes or not. By utilizing the scheme, on the basis of improving the accuracy of extracting the interaction keywords, the accuracy of representing the interaction content of the object resources by the interaction feature information is improved, and then the accuracy of identifying the merchant attributes is improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device and storage medium for object attribute recognition. Background Technology

[0002] With the widespread use of mobile payments, it is necessary to monitor and intercept suspicious resource interactions on online platforms. Analysis reveals that a significant proportion of those triggering suspicious behavior interception possess merchant attributes; therefore, identifying the merchant attributes of objects is particularly important.

[0003] In existing technologies, behavioral statistical analysis is usually performed on the resource interaction data of the object to be identified to obtain structured behavioral statistics (e.g., number of interactions, interaction amount, interaction method, etc.). Then, the model makes merchant attribute prediction based on the behavioral statistical features of the object to be identified. This not only easily leads to misjudgment, but also requires a large amount of labeled data for feature engineering processing of the model, resulting in high training costs. Summary of the Invention

[0004] This application provides an object attribute recognition method, apparatus, device, and storage medium. By collecting unstructured text data, it predicts merchant attributes from both object identification and interaction content dimensions. This improves the accuracy of interactive keyword extraction and enhances the accuracy of interactive feature information in representing the interactive content of object resources, thereby improving the accuracy of merchant attribute recognition. Furthermore, it eliminates the need for extensive labeled data for feature engineering, reducing training costs. The technical solution of this application is as follows:

[0005] On the one hand, an object attribute identification method is provided, the method comprising:

[0006] Obtain the identification text feature information of the object to be identified and multiple interaction record texts corresponding to the object to be identified. Each interaction record text is the associated text generated by the interactive object during a resource interaction with the object to be identified.

[0007] Interaction keywords are extracted from the multiple interaction record texts to obtain multiple interaction keywords;

[0008] The multiple interactive keywords are input into a keyword feature extraction model to extract keyword features and obtain interactive feature information.

[0009] Based on the identified text feature information and the interaction feature information, merchant attribute identification is performed on the object to be identified to obtain merchant attribute indication information, which is used to indicate whether the object to be identified has merchant attributes.

[0010] On the other hand, an object attribute recognition device is provided, the device comprising:

[0011] The data acquisition module is used to acquire the identification text feature information of the object to be identified and multiple interaction record texts corresponding to the object to be identified. Each interaction record text is the associated text generated by the interactive object during a resource interaction with the object to be identified.

[0012] The interactive keyword extraction module is used to extract interactive keywords from the multiple interactive record texts to obtain multiple interactive keywords;

[0013] The keyword feature extraction module is used to input the multiple interactive keywords into the keyword feature extraction model to extract keyword features and obtain interactive feature information.

[0014] The merchant attribute recognition module is used to identify the merchant attributes of the object to be identified based on the identifier text feature information and the interaction feature information, and to obtain merchant attribute indication information, which is used to indicate whether the object to be identified has merchant attributes.

[0015] On the other hand, an object attribute recognition device is provided, the device including a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the object attribute recognition method as described in the first aspect.

[0016] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction or at least one program is stored therein, the at least one instruction or the at least one program being loaded and executed by a processor to implement the object attribute identification method as described in the first aspect.

[0017] On the other hand, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the object attribute identification method as described in the first aspect.

[0018] The object attribute identification method, apparatus, device, and storage medium provided in this application have the following technical effects:

[0019] This application collects multiple interaction record texts generated during resource interactions between different interactive objects and the object to be identified. Interaction keywords are extracted from these texts to obtain multiple interaction keywords, and keyword features are extracted from these keywords to obtain interaction feature information. This improves the accuracy of interaction keyword extraction and enhances the accuracy of the interaction feature information in representing the content of object resource interactions. Based on the identifier text feature information and interaction feature information of the object to be identified, merchant attribute identification is performed. Compared to existing methods that predict merchant attributes based on structured behavioral statistical features, collecting unstructured text data to predict merchant attributes from both object identifier and interaction content dimensions not only improves the accuracy and interpretability of merchant attribute identification but also eliminates the need for extensive labeled data for model feature engineering, reducing training costs. Attached Figure Description

[0020] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of an application environment provided in an embodiment of this application;

[0022] Figure 2 This is a flowchart illustrating an object attribute identification method provided in an embodiment of this application;

[0023] Figure 3 This is a flowchart illustrating a process for extracting interactive keywords from multiple interactive record texts to obtain multiple interactive keywords, provided in an embodiment of this application.

[0024] Figure 4 This is a flowchart illustrating another process for extracting interactive keywords from multiple interactive record texts to obtain multiple interactive keywords, provided in an embodiment of this application.

[0025] Figure 5 This is a flowchart illustrating another object attribute identification method provided in an embodiment of this application;

[0026] Figure 6 This is a flowchart illustrating an interactive keyword extraction scheme provided in an embodiment of this application;

[0027] Figure 7This is a flowchart illustrating a process provided in this application embodiment of inputting multiple interactive keywords into a keyword feature extraction model to extract keyword features and obtain interactive feature information;

[0028] Figure 8 This is a flowchart illustrating a process for identifying merchant attributes of an object to be identified based on identifier text feature information and interaction feature information, and obtaining merchant attribute indication information, provided in an embodiment of this application.

[0029] Figure 9 This is a schematic diagram of a model structure provided in an embodiment of this application;

[0030] Figure 10 This is a block diagram of an object attribute recognition device provided in an embodiment of this application;

[0031] Figure 11 This is a schematic diagram of the structure of an object attribute recognition device provided in an embodiment of this application. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0033] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products or devices.

[0034] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0035] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0036] To facilitate understanding of the embodiments of this application, several concepts will be briefly introduced below:

[0037] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0038] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, pre-trained model technology, operating / interactive systems, and mechatronics. Among these, pre-trained models, also known as large-scale models or foundational models, can be widely applied to downstream tasks across various AI fields after fine-tuning. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0039] Natural Language Processing (NLP) is an important field within computer science and artificial intelligence. It studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP deals with natural language, the language people use in daily life, and is closely related to linguistics; it also involves computer science and mathematics. Pre-trained models, a crucial technique for model training in artificial intelligence, evolved from large language models in NLP. After fine-tuning, large language models can be widely applied to downstream tasks. NLP techniques typically include text processing, semantic understanding, machine translation, question answering, and knowledge graphs.

[0040] Pre-trained models, also known as foundational models or large models, refer to deep neural networks (DNNs) with a large number of parameters. These DNNs are trained on massive amounts of unlabeled data. Leveraging the function approximation capabilities of large-parameter DNNs, Proximity-Based Transformers (PTMs) extract common features from the data. Through fine-tuning, efficient parameter fine-tuning (PEFT), and prompt-tuning techniques, they are suitable for downstream tasks. Therefore, pre-trained models can achieve ideal results in small-shot or zero-shot scenarios. PTMs can be categorized according to the data modality they process, such as language models (ELMO, BERT, GPT), visual models (Swin-transformer, ViT, V-MOE), speech models (VALL-E), and multimodal models (ViBERT, CLIP, Flamingo, Gato). Multimodal models refer to models that establish feature representations for two or more data modalities. Pre-trained models are important tools for outputting AI-generated content (AIGC) and can also serve as a general interface connecting multiple task-specific models.

[0041] Machine Learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and pre-trained learning. Pre-trained models represent the latest development in deep learning, integrating all of these techniques.

[0042] With the research and advancement of artificial intelligence (AI) technology, AI is being studied and applied in various fields, such as smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, autonomous driving, drones, digital twins, virtual humans, robots, AI-generated content (AIGC), conversational interaction, smart healthcare, smart customer service, and game AI. It is believed that with the development of technology, AI will be applied in more fields and play an increasingly important role.

[0043] The object attribute identification method provided in this application embodiment can be applied to, for example... Figure 1In the application environment shown, the environment may include a client 10 and a server 20, which can be directly or indirectly connected via wired or wireless communication. The relevant object can send a merchant attribute identification request for the object to be identified to the server 20 through the client 10. Based on the merchant attribute identification request, the server 20 obtains the identifier text feature information of the object to be identified and multiple interaction record texts. Each interaction record text is associated text generated during a resource interaction between the interacting object and the object to be identified. Then, interaction keywords are extracted from the multiple interaction record texts to obtain multiple interaction keywords. These keywords are then input into a keyword feature extraction model for keyword feature extraction to obtain interaction feature information. Based on the identifier text feature information and interaction feature information, the server identifies the merchant attribute of the object to be identified, obtaining merchant attribute indication information. This merchant anomaly indication information is then fed back to the client 10. The merchant attribute indication information indicates whether the object to be identified has merchant attributes. Based on the merchant anomaly indication information, the client 10 performs resource interaction control on the merchant to be detected. It should be noted that... Figure 1 This is merely an example. The object attribute identification method provided in this application embodiment can be executed by the client or the server, or by both the client and the server. This application does not impose any restrictions on this.

[0044] The client can be a physical device such as a smartphone, computer (e.g., desktop computer, tablet computer, laptop computer), digital assistant, smart voice interaction device (e.g., smart speaker), smart wearable device, or software running on the physical device, such as a computer program. The operating system corresponding to the client can be Android, iOS (a mobile operating system developed by Apple), Linux (an operating system), Microsoft Windows, etc.

[0045] The server side can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The server may include network communication units, processors, and memory, etc. The server side can provide backend services to the corresponding clients.

[0046] The aforementioned client 10 and server 20 can be used to build a system for identifying object attributes, which can be a distributed system.

[0047] The following describes a specific embodiment of an object attribute recognition method provided in this application. Figure 2 This is a flowchart illustrating an object attribute identification method provided in an embodiment of this application. This application provides the operational steps of the method described in the embodiment or flowchart, but based on conventional or non-inventive labor, more or fewer operational steps may be included. The order of steps listed in the embodiment is merely one possible execution order among many and does not represent the only execution order. In actual systems or products, the method can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment) as shown in the embodiment or drawings. Specifically, as... Figure 2 As shown, the method may include:

[0048] S201, obtain the identification text feature information of the object to be identified and multiple interaction record texts corresponding to the object to be identified. Each interaction record text is the associated text generated by the interactive object during a resource interaction with the object to be identified.

[0049] In this illustrative embodiment, the object to be identified can be a business object on a resource interaction platform. This resource interaction platform may include, but is not limited to, banking platforms, shopping platforms, and instant messaging platforms. The business object can refer to a user or a user account; illustratively, the user account may include, but is not limited to, an account for storing resources or transferring resources through it, such as a bank card account opened at a bank, a funds account opened in the account system of a shopping platform, and a funds account opened in the account system of an instant messaging platform.

[0050] In this embodiment, the identifier text feature information of the object to be identified can be used to characterize the semantic features of the object identifier text of the object to be identified. In a specific embodiment, the identifier text feature information can be represented as an identifier text feature vector.

[0051] In one specific embodiment, obtaining the identifier text feature information of the object to be identified may include:

[0052] S2011, Obtain the object identifier text of the object to be identified;

[0053] S2012, input the object identification text into the identification text feature extraction model to extract text features and obtain identification text feature information.

[0054] Specifically, object identification text can be descriptive text that identifies the object to be identified. Indicatively, object identification text may include, but is not limited to: object name text, object nickname text, object description text, etc.

[0055] Specifically, the identifier text feature extraction model can be used to extract text features from object identifier text. The identifier text feature extraction model can be any existing artificial intelligence model with text feature extraction capabilities; this application does not impose any particular limitation on it. Schematic, the identifier text feature extraction model can adopt a bidirectional Transformer structure.

[0056] In the embodiments of this specification, each interaction record text can be associated text generated by the interactive object during a resource interaction with the object to be identified. Specifically, the resources may include, but are not limited to, monetary assets used for online transactions, such as real resources like RMB, and virtual resources like game coins and game points; correspondingly, resource interaction may include actions such as resource transfer and payment, such as online payment, bank transfer, and sending electronic red envelopes.

[0057] Specifically, the interaction record text is the interaction-related text generated by the interaction object of the object to be identified. In a specific embodiment, the interaction object here can refer to the interaction initiating object, such as the object that initiates a transfer to the object to be identified, or the object that sends a red envelope to the object to be identified. In optional embodiments, multiple interaction record texts can correspond to different interaction objects, or they can correspond to the same interaction object.

[0058] In one specific embodiment, the interaction record text may include, but is not limited to, interaction notes and interaction feedback text. Specifically, the interaction notes can be notes generated when the interaction object initiates a resource interaction, such as transfer notes, red envelope text, etc.; the interaction feedback text can be feedback text generated when the interaction object completes the resource interaction, such as complaint text, etc.

[0059] S202, Extract interaction keywords from multiple interaction record texts to obtain multiple interaction keywords.

[0060] In the embodiments of this specification, interaction keywords can be used to express the resource interaction behavior content of the object to be identified. Specifically, interaction keywords can be important words that reflect the exchange content provided by the object to be identified to the interaction object.

[0061] In one specific embodiment, the aforementioned multiple interaction record texts can be selected from all interaction record texts corresponding to the object to be identified, with their generation time falling within a target time period. The target time period may include multiple sub-time periods. Specifically, the setting of the target time period and the division of sub-time periods can be pre-set based on the interaction record text classification requirements in actual applications.

[0062] In an optional embodiment, each of the multiple sub-time periods can be a continuous time interval. Schematically, taking the target time period as the A year, the multiple sub-time periods can be multiple months within the A year. For example, the multiple sub-time periods can include: January of year A, February of year A... December of year A.

[0063] In an optional embodiment, each of the multiple sub-time periods can be a periodic discrete time interval. Schematically, taking the target time period as month B, the multiple sub-time periods can include: Monday time period, Tuesday time period... Sunday time period. Specifically, the Monday time period can include multiple Mondays within month B, the Tuesday time period can include multiple Tuesdays within month B... the Sunday time period can include multiple Sundays within month B.

[0064] In an optional embodiment, as Figure 3 shown, the generation time period of the above-mentioned multiple interactive record texts can include multiple sub-time periods. The extraction of multiple interactive keywords from the above-mentioned multiple interactive record texts can include:

[0065] S301, perform word segmentation on at least one target text generated within a target sub-time period to obtain multiple text segments; the target sub-time period is any one of the multiple sub-time periods, and at least one target text is at least one interactive record text generated within the target sub-time in the multiple interactive record texts.

[0066] Specifically, perform word segmentation on at least one target text according to punctuation marks (such as full stops, question marks, exclamation marks, commas, etc.) and preset stop words (such as "de", "shi", etc.) to obtain multiple text segments.

[0067] In an optional embodiment, before performing word segmentation on at least one target text, noise reduction processing can be performed on at least one target text first. Specifically, the noise reduction processing can include: 1) removing noise texts such as spaces, illegal characters, emoticons, numeric texts (such as numbers, mobile phone numbers, etc.); 2) removing texts whose lengths do not meet the preset constraint lengths.

[0068] S302, determine the lexical proportion index of each text segment in at least one target text.

[0069] Specifically, the lexical proportion index of each text segment can represent the proportion of the corresponding text segment in at least one target text.

[0070] In one specific embodiment, the vocabulary proportion index of each text segment can be the ratio between the first word frequency of each text segment and the total number of segments of at least one target text, wherein the first word frequency of each text segment can be the number of times each text segment appears in the at least one target text.

[0071] S303, determine the time period ratio of the coverage sub-time period corresponding to each text segment in multiple sub-time periods. The coverage sub-time period corresponding to each text segment is the sub-time period in which at least one target text generated in multiple sub-time periods contains the corresponding text segment.

[0072] Specifically, the time period percentage index corresponding to each text segment can characterize the distribution of the corresponding text segment in multiple interactive records.

[0073] In a specific embodiment, the time period ratio index corresponding to each text segment can be the ratio between the number of covered sub-time periods corresponding to each text segment and the total number of sub-time periods within the generated time period.

[0074] S304, based on the vocabulary proportion index and the time period proportion index, performs text importance analysis on each text segment for the target sub-time period to obtain the first importance representation index of each text segment.

[0075] Specifically, the first importance metric for each text segment can characterize the importance of the corresponding text segment to at least one target text generated within the target sub-time period.

[0076] In an optional embodiment, the ratio between the vocabulary proportion index and the time period proportion index of each text segment can be used as the first importance index of each text segment, or the ratio between the vocabulary proportion index and the target logarithm (the logarithm of the time period proportion index) of each text segment can be used as the first importance index of each text segment.

[0077] Understandably, the larger the value of the lexical proportion index of a certain text segment, the higher the frequency of occurrence of that text segment in at least one target text generated within the target sub-time period. Conversely, the smaller the value of the time period proportion index of a certain text segment, the stronger the representativeness of that text segment to the text content generated within the target sub-time period compared to other sub-time periods. Therefore, based on the ratio between the lexical proportion index and the time period proportion index of each text segment, the first importance representation index of each text segment is obtained. The larger the value of the first importance representation index of a certain text segment, the higher the importance of that text segment to at least one target text generated within the target sub-time period.

[0078] S305, determine at least one initial keyword corresponding to the target sub - time period from multiple text segments based on the first importance representation index of each of the multiple text segments.

[0079] In a specific embodiment, the multiple text segments can be sorted based on the descending order of the corresponding values of the first importance representation index, and the top N text segments in the sorting are used as the initial keywords corresponding to the target sub - time period, where N≥1.

[0080] S306, use at least one initial keyword corresponding to each of the multiple sub - time periods as multiple interaction keywords.

[0081] As can be seen from the above embodiments, since the resource interaction behaviors and resource interaction contents of the object to be recognized may be different in different time periods, by evaluating the first importance representation index of each text segment in a specific sub - time period, extracting the text segments that frequently appear in a specific sub - time period and rarely appear in other sub - time periods as the initial keywords of the specific sub - time period, the interaction content features of the object to be recognized in different time periods can be better extracted.

[0082] In an optional embodiment, as Figure 4 shown, the extraction of multiple interaction keywords from multiple interaction record texts can include:

[0083] S401, perform word - segmenting processing on multiple interaction record texts to obtain multiple candidate phrases and at least one word segment included in each candidate phrase.

[0084] Specifically, according to punctuation marks (such as full stops, question marks, exclamation marks, commas, etc.) and preset stop words (such as "de", "shi", etc.), perform word - segmenting processing on multiple interaction record texts to obtain multiple candidate phrases, and based on the spaces in each candidate phrase, split out at least one word segment included in each candidate phrase.

[0085] In an optional embodiment, before performing word - segmenting processing on multiple interaction record texts, noise reduction processing can be performed on the multiple interaction record texts first. Specifically, the noise reduction processing can include: 1) removing noise texts such as illegal characters, emoticons, numerical texts (such as numbers, mobile phone numbers, etc.); 2) removing texts whose lengths do not meet the preset constraint lengths.

[0086] S402, perform context influence analysis on at least one word segment included in each candidate phrase respectively to obtain the second importance representation index of each candidate phrase.

[0087] Specifically, the second importance representation index of each candidate phrase can represent the importance of the corresponding candidate phrase for multiple interaction record texts.

[0088] In a specific embodiment, the above-described contextual influence analysis of at least one word segment contained in each candidate phrase, to obtain a second importance representation index for each candidate phrase, may include:

[0089] S4021, determine the number of times each segmentation contained in each candidate phrase co-occurs in multiple candidate phrases.

[0090] Specifically, the co-occurrence count of each word can refer to the number of times each word co-occurs with other words in multiple candidate phrases. Here, "other words" can be any word among all the words contained in multiple candidate phrases.

[0091] In an optional embodiment, the co-occurrence count of a word is incremented by 1 each time it co-occurs with another word in a candidate phrase. It can be understood that the co-occurrence count of a word with itself can be the number of times the word appears in multiple candidate phrases, i.e., word frequency.

[0092] S4022, determine the number of times each word appears in multiple candidate phrases.

[0093] Specifically, the number of occurrences of each word can refer to the word frequency of each word among all the words contained in multiple candidate phrases.

[0094] S4023, determine the contextual influence index of each word based on the ratio between the co-occurrence frequency of each word and the occurrence frequency of each word.

[0095] Specifically, the contextual impact index of each word can characterize the importance of the corresponding word to multiple interactive record texts. In an optional embodiment, the ratio between the co-occurrence frequency of each word and the occurrence frequency of each word can be used as the contextual impact index of each word. Generally, the larger the value of the contextual impact index of a word, the higher the importance of that word to multiple interactive record texts.

[0096] S4024, perform index fusion on the contextual influence index of at least one word contained in each candidate phrase to obtain the second importance characterization index of each candidate phrase.

[0097] In an optional embodiment, the sum of the contextual influence metrics of at least one word contained in each candidate phrase can be used as the second importance metric for each candidate phrase; alternatively, the mean of the contextual influence metrics of at least one word contained in each candidate phrase can be used as the second importance metric for each candidate phrase.

[0098] S403, based on the second importance representation index of each of the multiple candidate phrases, determines multiple interactive keywords from the multiple candidate phrases.

[0099] In a specific embodiment, multiple candidate phrases can be sorted based on the numerical values ​​of the second importance indicator from largest to smallest, and the top N2 candidate phrases in the sorting can be used as multiple interactive keywords, where N2≥2.

[0100] To illustrate, taking the interactive record text "The color of the skirt is different from the seller's photos, and the skirt is of poor quality" as an example, we perform word segmentation on this interactive record text to obtain candidate phrase 1: 'skirt', candidate phrase 2: 'color', candidate phrase 3: 'seller's photos', candidate phrase 4: 'color difference', and candidate phrase 5: 'the skirt is of poor quality'. Further breaking down the 5 candidate phrases, we get 6 word segments: 'skirt', 'color', 'seller's photos', 'color difference', 'quality', and 'poor'.

[0101] Among them, word segment 1 'skirt' co-occurred 4 times and appeared 2 times in the 5 candidate phrases, with a contextual influence index of 2; word segment 2 'color' co-occurred 1 time and appeared 1 time in the 5 candidate phrases, with a contextual influence index of 1; word segment 3 'seller's show' co-occurred 1 time and appeared 1 time in the 5 candidate phrases, with a contextual influence index of 1; word segment 4 'color difference' co-occurred 1 time and appeared 1 time in the 5 candidate phrases, with a contextual influence index of 1; word segment 5 'quality' co-occurred 3 times and appeared 1 time in the 5 candidate phrases, with a contextual influence index of 3; word segment 6 'bad' co-occurred 3 times and appeared 1 time in the 5 candidate phrases, with a contextual influence index of 3.

[0102] Correspondingly, the second importance index of candidate phrase 1 is 2, the second importance index of candidate phrase 2 is 1, the second importance index of candidate phrase 3 is 1, the second importance index of candidate phrase 4 is 1, and the second importance index of candidate phrase 5 is 8. Based on the order of the values ​​corresponding to the second importance index from largest to smallest, the five candidate phrases are sorted, and the sorting results are: 'The skirt is of poor quality', 'Skirt', 'Color', 'Seller's show', 'Color difference'.

[0103] As can be seen from the above embodiments, by performing word segmentation on multiple interactive record texts, multiple candidate phrases and at least one word contained in each candidate phrase are obtained. By evaluating the contextual influence index of each candidate phrase and at least one word contained in each candidate phrase, a second importance representation index of each candidate phrase is obtained, thereby screening out multiple interactive keywords and improving the accuracy of interactive keyword extraction.

[0104] In an optional embodiment, such as Figure 5 As shown, the above-mentioned extraction of interaction keywords from multiple interaction record texts can also yield multiple interaction keywords, including:

[0105] S2021, Extract interaction keywords from multiple interaction record texts to obtain multiple initial keywords.

[0106] Specifically, the detailed steps of "extracting interactive keywords from multiple interactive record texts to obtain multiple initial keywords" can be found in the detailed steps S301 to S305, or in the detailed steps S401 to S403, and will not be repeated here.

[0107] S2022, multiple initial keywords are filtered by part of speech to obtain multiple interactive keywords.

[0108] Specifically, in resource interaction scenarios, keywords that can reflect the interactive content are generally composed of nouns or pronouns. Therefore, it is necessary to perform part-of-speech parsing on multiple initial keywords to obtain the part of speech of each initial keyword, and then use the initial keywords with the corresponding part of speech as nouns or pronouns as interactive keywords.

[0109] In an optional embodiment, when the initial keyword is a phrase composed of multiple word segments, the part-of-speech tagging of each word segment in the initial keyword can be performed to obtain the part-of-speech tag of each word. Then, the word segments in the initial keyword that correspond to nouns or pronouns are combined to obtain the interactive keyword corresponding to the initial keyword. For example, taking the initial keyword "skirt quality is poor," the word segment "skirt" has a noun part of speech, the word segment "quality" has a noun part of speech, and the word segment "poor" has an adjective part of speech. Therefore, combining the word segments "skirt" and "quality" yields the interactive keyword "skirt quality."

[0110] illustrative, see Figure 6 , Figure 6 This is a flowchart illustrating an interactive keyword extraction scheme provided in an embodiment of this application. Specifically:

[0111] First step, select multiple interaction record texts of the object to be recognized within the target time for splicing to obtain an interaction splicing text. Perform word segmentation on the multiple interaction record texts according to punctuation marks (such as full stops, question marks, exclamation marks, commas, etc.) and preset stop words (such as "de", "shi", etc.) to obtain multiple candidate phrases, and based on the spaces in each candidate phrase, split out at least one word segment included in each candidate phrase; then determine the context influence index corresponding to each word segment. Specifically, the context influence index wordScore(w) of word segment w = wordDegree(w) / wordFrequency(w), where wordDegree(w) represents the co-occurrence times of word segment w in multiple candidate phrases, and wordFrequency(w) represents the co-occurrence times of word segment w in multiple candidate phrases, and perform index fusion on the context influence indexes of at least one word segment included in each candidate phrase to obtain the second importance characterization index of each candidate phrase.

[0112] Second step, sort the multiple candidate phrases according to the numerical values of the second importance characterization indexes and output them in descending order. Since the interaction keywords in the resource interaction scenario are generally composed of nouns or pronouns, it is necessary to perform word property parsing on the candidate phrases to obtain the word properties of each word segment in the candidate phrases, and then filter and combine the word segments with corresponding word properties of nouns or pronouns to obtain the interaction keywords corresponding to the candidate phrases.

[0113] Third step, if the filtering result after word property filtering is not a null value, then select the first K interaction keywords as the target result for output from the sorted interaction keywords. Illustratively, K can be taken as 100 here; if the filtering result after word property filtering is a null value, directly perform the word property filtering process in the second step on the interaction splicing text obtained in the first step, and finally take the first 100 word segments in reverse order as the target result for output.

[0114] S203, input the multiple interaction keywords into a keyword feature extraction model to extract keyword features and obtain interaction feature information.

[0115] As can be seen from the above embodiments, through noise reduction and key information extraction, and further performing word property parsing and word property filtering, better keyword extraction effects can be obtained, and the accuracy of keyword extraction can be improved.

[0116] In the embodiments of this specification, the interaction feature information can be used to characterize the interaction content features of the object to be recognized during the resource interaction process. Specifically, the interaction feature information can be obtained by inputting the multiple interaction keywords into a keyword feature extraction model for keyword feature extraction. In a specific embodiment, the manifestation form of the interaction feature information can be an interaction feature vector.

[0117] Specifically, the keyword feature extraction model can be used to extract keyword features from multiple interactive keywords. The keyword feature extraction model can be any existing artificial intelligence model with keyword feature extraction capabilities. For example, the keyword feature extraction model can adopt a bidirectional Transformer structure.

[0118] In an optional embodiment, the time period for generating the aforementioned multiple interactive record texts may include multiple sub-time periods, and the aforementioned multiple interactive keywords may include at least one initial keyword corresponding to each of the multiple sub-time periods. The aforementioned keyword feature extraction model may include: a word segmentation semantic extraction layer, a time period text semantic extraction layer, a temporal feature extraction layer, and a feature aggregation layer, such as... Figure 7 As shown, the above-mentioned input of multiple interactive keywords into the keyword feature extraction model for keyword feature extraction yields interactive feature information that may include:

[0119] S701, according to the time sequence of multiple sub-time periods, concatenate at least one initial keyword corresponding to each of the multiple sub-time periods to obtain the target concatenated text.

[0120] Specifically, you can first concatenate at least one initial keyword corresponding to each sub-time period to obtain the time period text corresponding to each sub-time period. Then, according to the time order of multiple sub-time periods, concatenate the time period texts corresponding to each of the multiple sub-time periods to obtain the target concatenated text.

[0121] In an optional embodiment, concatenating at least one initial keyword corresponding to each sub-time period can refer to concatenating them in order of the numerical value of the first importance indicator of each initial keyword.

[0122] In an optional embodiment, a preset identifier can be added to the beginning of the target concatenated text. This preset identifier can be used to help the model capture global information describing the word segmentation sequence. Illustratively, the preset identifier can be represented as "cls".

[0123] S702, input the target concatenated text into the word segmentation semantic extraction layer, perform word segmentation semantic extraction on at least one initial keyword corresponding to multiple sub-time periods, and obtain word segmentation semantic feature information corresponding to each initial keyword.

[0124] Specifically, the word segmentation semantic feature information corresponding to each initial keyword can represent the semantic features of the corresponding initial keyword itself. In a specific embodiment, the word segmentation semantic feature information corresponding to each initial keyword can be represented as a word segmentation semantic feature vector.

[0125] Specifically, the word segmentation semantic extraction layer can be used to perform word segmentation semantic extraction on each initial keyword in the target concatenated text. Schematically, the model structure of the word segmentation semantic extraction layer can adopt a text encoder.

[0126] S703. Input the target concatenated text into the time segment text semantic extraction layer, perform semantic extraction on at least one initial keyword corresponding to the sub-time segment to which each initial keyword belongs, and obtain the time segment text feature information corresponding to each initial keyword.

[0127] Specifically, the time segment text feature information corresponding to each initial keyword can represent the semantic features of the time segment text corresponding to the sub-time segment to which the corresponding initial keyword belongs. The time segment text corresponding to the sub-time segment here can be obtained by concatenating at least one initial keyword corresponding to the sub-time segment. In a specific embodiment, the manifestation form of the time segment text feature information corresponding to each initial keyword can be a time segment text feature vector.

[0128] Specifically, the time segment text semantic extraction layer can be used to perform text semantic extraction on the time segment texts of multiple sub-time segments in the target concatenated text respectively. Schematically, the model structure of the time segment text semantic extraction layer can adopt a text encoder.

[0129] S704. Input the target concatenated text into the time sequence feature extraction layer, perform time sequence feature extraction on the sub-time segments to which each initial keyword belongs, and obtain the time sequence feature information corresponding to each initial keyword.

[0130] Specifically, the time sequence feature information corresponding to each initial keyword can represent the sequential feature of the sub-time segment to which the corresponding initial keyword belongs. In a specific embodiment, the manifestation form of the time sequence feature information corresponding to each initial keyword can be a time sequence feature vector.

[0131] Specifically, the time sequence feature extraction layer can be used to perform position encoding processing on the sequential relationship of multiple sub-time segments corresponding to the target concatenated text.

[0132] In an optional embodiment, the time sequence feature extraction layer can perform static encoding processing. Schematically, the time sequence feature vector corresponding to the j-th sub-time segment to which each initial keyword belongs can be expressed as: Pj = [e(j,1), e(j,2), …, e(j,2i), e(j,2i + 1), …, e(j,d)], where, d represents the dimension size of the vector space. Optionally, d can be 128; n represents a custom scalar. Optionally, n can be 10000; i is used to map to the column index of the vector element, 0 ≤ i < d / 2, and a single value of i is mapped to both the sine function and the cosine function.

[0133] In an optional embodiment, the temporal feature extraction layer can perform dynamic encoding processing, generating a dynamic temporal feature vector for each sub-time period by calculating the relative or absolute time distance between different sub-time periods.

[0134] S705 inputs word segmentation semantic feature information, time period text feature information and time sequence feature information into the feature aggregation layer for contextual semantic fusion to obtain interactive feature information.

[0135] In one specific embodiment, the feature aggregation layer may include a bidirectional semantic encoder, which inputs word segmentation semantic feature information, time period text feature information and time sequence feature information into the bidirectional semantic encoder, and performs contextual semantic aggregation through the bidirectional semantic encoder to obtain interactive feature information corresponding to preset identifiers.

[0136] For illustrative purposes, the bidirectional semantic encoder here can be an encoder with a bidirectional Transformer structure.

[0137] As can be seen from the above embodiments, according to the time sequence of multiple sub-time periods, at least one initial keyword corresponding to each of the multiple sub-time periods is concatenated to obtain the target concatenated text. The target concatenated text is then input into the word segmentation semantic extraction layer, the time period text semantic extraction layer, and the temporal feature extraction layer for feature extraction, respectively, to obtain the word segmentation semantic feature information, the time period text feature information, and the temporal feature information corresponding to each initial keyword. These three features are then fused with contextual semantics, which can more fully integrate the interactive content features of the object to be identified in different time periods, thereby improving the accuracy of the interactive feature information in representing the resource interactive content of the object to be identified.

[0138] S204. Based on the identifier text feature information and interaction feature information, perform merchant attribute identification on the object to be identified to obtain merchant attribute indication information. The merchant attribute indication information is used to indicate whether the object to be identified has merchant attributes.

[0139] In the embodiments of this specification, merchant attribute indication information can be used to indicate whether the object to be identified has merchant attributes, where merchant attributes can refer to whether the object to be identified has commercial business activities.

[0140] In one specific embodiment, the merchant attribute indication information may include: attribute prediction probability information, which can be used to characterize the predicted probability that the object to be identified has merchant attributes.

[0141] In a specific embodiment, such as Figure 8 As shown, the merchant attribute identification information obtained by identifying the object to be identified based on the above-mentioned text feature information and interaction feature information may include:

[0142] S2041, input the identifier text feature information and interaction feature information into the feature fusion model for feature fusion processing to obtain the target attribute feature information.

[0143] Specifically, target attribute feature information can be used to characterize the commercial attribute features of the object to be identified. This target attribute feature information can be obtained by fusing identifier text feature information and interaction feature information. In a specific embodiment, the target attribute feature information can be represented as a target attribute feature vector.

[0144] In an optional embodiment, the feature fusion model may include a pooling layer and a concatenation layer. Specifically, the pooling layer may be used to perform average pooling on the identifier text feature information and the interaction feature information respectively, and the concatenation layer may be used to concatenate the pooled feature information corresponding to the identifier text feature information and the pooled feature information corresponding to the interaction feature information.

[0145] S2042, Input the target attribute feature information into the merchant attribute recognition model for attribute recognition to obtain merchant attribute indication information.

[0146] Specifically, the merchant attribute recognition model can be used to identify attributes based on target attribute feature information. In an optional embodiment, the merchant attribute recognition model can adopt an MLP (Multilayer Perceptron). Schematic, the merchant attribute recognition model can include an input layer, a hidden layer, and an output layer. The activation function of the hidden layer can be the ReLU function, which can introduce nonlinear factors. The merchant attribute recognition model can also adopt other existing artificial intelligence models with classification and prediction capabilities; this application does not impose any particular limitations on this.

[0147] As can be seen from the above embodiments, by performing feature fusion processing on the identifier text feature information and interaction feature information to obtain target attribute feature information, and by performing merchant attribute identification from the object identifier dimension and interaction content dimension based on the target attribute feature information, the accuracy and interpretability of the merchant attribute identification results can be improved.

[0148] In one specific embodiment, see Figure 9 A merchant identification model can be constructed that includes the above-mentioned identifier text feature extraction model, keyword feature extraction model, feature fusion model and merchant attribute identification model. The object identifier text of the merchant to be identified and multiple interaction keywords from multiple interaction record texts are input into the merchant identification model to obtain the merchant attribute indication information of the merchant to be identified.

[0149] In one specific embodiment, the above method may further include:

[0150] S901, obtain the merchant attribute annotation information of the sample object, the sample object identification text of the sample object, and multiple sample interaction keywords of the sample object.

[0151] Specifically, the sample object can be a platform history object used for model training.

[0152] In practical applications, training data can be determined before network training. Specifically, in this embodiment, sample object data containing merchant attribute annotation information can be obtained as training data. Specifically, the merchant attribute annotation information can be pre-annotated merchant attribute tags for the sample objects. Multiple sample interaction keywords for the sample objects can be obtained by extracting interaction keywords from the corresponding multiple sample interaction record texts of the sample objects.

[0153] S902, input the sample object identification text into the preset text feature extraction model to extract text features and obtain the sample identification text feature information.

[0154] S903: Input multiple sample interaction keywords into a preset keyword feature extraction model to extract keyword features and obtain sample interaction feature information.

[0155] S904. Input the sample identification text feature information and sample interaction feature information into the preset feature fusion model for feature fusion processing to obtain sample attribute feature information.

[0156] S905, input the sample attribute feature information into the preset merchant attribute recognition model to perform attribute recognition and obtain merchant attribute prediction information.

[0157] S906 determines the target loss information based on merchant attribute labeling information and merchant attribute prediction information.

[0158] In an optional embodiment, the target loss information may include industry category distribution loss;

[0159] Accordingly, the above-mentioned determination of target loss information based on merchant attribute labeling information and merchant attribute prediction information may include: determining the merchant attribute distribution loss based on merchant attribute labeling information and merchant attribute prediction information.

[0160] In a specific embodiment, determining the merchant attribute distribution loss based on the merchant attribute annotation information and the merchant attribute prediction information may include determining the merchant attribute distribution loss between the merchant attribute annotation information and the merchant attribute prediction information based on a preset loss function.

[0161] In a specific embodiment, the merchant attribute distribution loss can characterize the difference between merchant attribute annotation information and merchant attribute prediction information.

[0162] In a specific embodiment, the preset loss function may include, but is not limited to, the cross-entropy loss function, the logistic loss function, the exponential loss function, etc.

[0163] S907, based on target loss information, train a preset text feature extraction model, a preset keyword feature extraction model, a preset feature fusion model and a preset merchant attribute recognition model to obtain the identifier text feature extraction model, keyword feature extraction model, feature fusion model and merchant attribute recognition model.

[0164] In an optional embodiment, based on target loss information, a preset text feature extraction model, a preset keyword feature extraction model, a preset feature fusion model, and a preset merchant attribute recognition model are trained to obtain the identifier text feature extraction model, keyword feature extraction model, feature fusion model, and merchant attribute recognition model, which may include:

[0165] S9071, based on target loss information, update the network parameters of the preset text feature extraction model, the preset keyword feature extraction model, the preset feature fusion model, and the preset merchant attribute recognition model;

[0166] S9072, based on the updated preset text feature extraction model, the updated preset keyword feature extraction model, the updated preset feature fusion model, and the updated preset merchant attribute recognition model, repeat the merchant attribute recognition training iteration operation including steps S902-S906 and S9071 until the merchant attribute recognition convergence condition is met; use the preset text feature extraction model obtained when the merchant attribute recognition convergence condition is met as the identifier text feature extraction model; use the preset keyword feature extraction model obtained when the merchant attribute recognition convergence condition is met as the keyword feature extraction model; use the preset feature fusion model obtained when the merchant attribute recognition convergence condition is met as the feature fusion model; use the preset merchant attribute recognition model obtained when the merchant attribute recognition convergence condition is met as the merchant attribute recognition model.

[0167] In an optional embodiment, the convergence condition for merchant attribute recognition can be that the number of training iterations reaches a preset number of training iterations. Optionally, the convergence condition for merchant attribute recognition can also be that the target loss information is less than a specified threshold. In the embodiments of this specification, the preset number of training iterations and the specified threshold can be preset in conjunction with the training speed and accuracy of the network in practical applications.

[0168] As can be seen from the technical solutions provided in the embodiments of this application above, by collecting multiple interaction record texts generated during resource interaction between different interactive objects and the object to be identified, extracting interaction keywords from these multiple interaction record texts to obtain multiple interaction keywords, and extracting keyword features from these multiple interaction keywords to obtain interaction feature information, the accuracy of interaction keyword extraction can be improved, as well as the accuracy of the interaction feature information in representing the content of object resource interaction. Furthermore, based on the identifier text feature information and interaction feature information of the object to be identified, merchant attribute identification can be performed on the object to be identified. Compared with the existing method of predicting merchant attributes based on structured behavioral statistical features, collecting unstructured text data to predict merchant attributes from the dimensions of object identifier and interaction content not only improves the accuracy and interpretability of merchant attribute identification, but also eliminates the need for a large amount of labeled data for model feature engineering, thus reducing training costs.

[0169] This application also provides an object attribute recognition device, such as... Figure 10 As shown, the object attribute recognition device may include:

[0170] The data acquisition module 1010 is used to acquire the identification text feature information of the object to be identified and multiple interaction record texts corresponding to the object to be identified. Each interaction record text is the associated text generated by the interactive object during a resource interaction with the object to be identified.

[0171] The interactive keyword extraction module 1020 is used to extract interactive keywords from multiple interactive record texts to obtain multiple interactive keywords;

[0172] The keyword feature extraction module 1030 is used to input multiple interactive keywords into the keyword feature extraction model to extract keyword features and obtain interactive feature information.

[0173] The merchant attribute recognition module 1040 is used to recognize the merchant attributes of the object to be identified based on the identifier text feature information and interaction feature information, and obtain the merchant attribute indication information. The merchant attribute indication information is used to indicate whether the object to be identified has merchant attributes.

[0174] In an optional embodiment, the time period for generating the aforementioned multiple interaction record texts may include multiple sub-time periods, and the aforementioned interaction keyword extraction module 1020 may include:

[0175] The first word segmentation unit is used to segment at least one target text generated within the target sub-time period to obtain multiple text segments; the target sub-time period is any sub-time period among multiple sub-time periods, and at least one target text is at least one interactive record text generated within the target sub-time period among multiple interactive record texts;

[0176] The vocabulary proportion indicator determination unit is used to determine the vocabulary proportion indicator of each text segment in at least one target text.

[0177] The time period percentage indicator unit is used to determine the time period percentage of each text segmentation corresponding to the covered sub-time period in multiple sub-time periods. The covered sub-time period corresponding to each text segmentation is the sub-time period in which at least one target text generated in multiple sub-time periods contains the corresponding text segmentation.

[0178] The first importance representation index unit is used to perform text importance analysis for each text segment based on the word proportion index and the time period proportion index, and obtain the first importance representation index for each text segment.

[0179] The initial keyword determination unit is used to determine at least one initial keyword corresponding to the target sub-time period from multiple text segments based on the first importance representation index of each of the multiple text segments;

[0180] The first interactive keyword determination unit is used to determine at least one initial keyword corresponding to each of the multiple sub-time periods as multiple interactive keywords.

[0181] In an optional embodiment, the interactive keyword extraction module 1020 may include:

[0182] The second word segmentation unit is used to segment multiple interactive record texts to obtain multiple candidate phrases and at least one word contained in each candidate phrase;

[0183] The second importance characterization index unit is used to perform contextual influence analysis on at least one word contained in each candidate phrase to obtain the second importance characterization index of each candidate phrase.

[0184] The second interactive keyword determination unit is used to determine multiple interactive keywords from multiple candidate phrases based on the second importance representation index of each candidate phrase.

[0185] In one specific embodiment, the above-mentioned second importance characterization index unit may include:

[0186] The co-occurrence frequency determination unit is used to determine the number of times each word contained in each candidate phrase co-occurs in multiple candidate phrases;

[0187] The occurrence count determination unit is used to determine the occurrence count of each word in multiple candidate phrases;

[0188] The context influence index determination unit is used to determine the context influence index of each word based on the ratio between the co-occurrence frequency of each word and the occurrence frequency of each word.

[0189] The indicator fusion unit is used to fuse the contextual influence indicators of at least one word contained in each candidate phrase to obtain a second importance representation indicator for each candidate phrase.

[0190] In an optional embodiment, the interactive keyword extraction module 1020 may further include:

[0191] The keyword extraction unit is used to extract interactive keywords from multiple interactive record texts to obtain multiple initial keywords.

[0192] The part-of-speech filtering unit is used to filter multiple initial keywords by part of speech to obtain multiple interactive keywords.

[0193] In an optional embodiment, the time period for generating the aforementioned multiple interactive record texts may include multiple sub-time periods, and the aforementioned multiple interactive keywords may include at least one initial keyword corresponding to each of the multiple sub-time periods. The aforementioned keyword feature extraction model may include: a word segmentation semantic extraction layer, a time period text semantic extraction layer, a temporal feature extraction layer, and a feature aggregation layer. The aforementioned keyword feature extraction module 1030 may include:

[0194] The keyword concatenation unit is used to concatenate at least one initial keyword corresponding to each of the multiple sub-time periods according to the time sequence of the sub-time periods to obtain the target concatenated text.

[0195] The word segmentation semantic extraction unit is used to input the target concatenated text into the word segmentation semantic extraction layer, and perform word segmentation semantic extraction on at least one initial keyword corresponding to multiple sub-time periods to obtain word segmentation semantic feature information corresponding to each initial keyword.

[0196] The time-segment text semantic extraction unit is used to input the target concatenated text into the time-segment text semantic extraction layer, perform semantic extraction on at least one initial keyword corresponding to the sub-time period to which each initial keyword belongs, and obtain the time-segment text feature information corresponding to each initial keyword;

[0197] The temporal feature extraction unit is used to input the target concatenated text into the temporal feature extraction layer, extract temporal features for the sub-time period to which each initial keyword belongs, and obtain the temporal feature information corresponding to each initial keyword;

[0198] The context semantic fusion unit is used to input word segmentation semantic feature information, time period text feature information and time sequence feature information into the feature aggregation layer for context semantic fusion to obtain interactive feature information.

[0199] In one specific embodiment, the merchant attribute identification module 1040 described above may include:

[0200] The feature fusion unit is used to input the identifier text feature information and interaction feature information into the feature fusion model for feature fusion processing to obtain the target attribute feature information;

[0201] The attribute recognition unit is used to input the target attribute feature information into the merchant attribute recognition model for attribute recognition and obtain merchant attribute indication information.

[0202] It should be noted that the apparatus and method embodiments described in the device embodiments are based on the same inventive concept.

[0203] This application provides an object attribute recognition device, which includes a processor and a memory. The memory stores at least one instruction or at least one program segment, which is loaded and executed by the processor to implement the object attribute recognition method provided in the above method embodiments.

[0204] Furthermore, Figure 11 A schematic diagram of the hardware structure of an object attribute recognition device for implementing the object attribute recognition method provided in the embodiments of this application is shown. The object attribute recognition device may participate in or include the object attribute recognition apparatus provided in the embodiments of this application. Figure 11 As shown, the object attribute recognition device 110 may include one or more processors 1102 (shown as 1102a, 1102b, ..., 1102n in the figure) (processor 1102 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 1104 for storing data, and a transmission device 1106 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 11 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, the object attribute recognition device 110 may also include a... Figure 11 The more or fewer components shown, or having the same Figure 11 The different configurations shown.

[0205] It should be noted that the aforementioned one or more processors 1102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the object attribute identification device 110 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0206] The memory 1104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the object attribute recognition method described in the embodiments of this application. The processor 1102 executes various functional applications and data processing by running the software programs and modules stored in the memory 1104, thereby realizing the above-mentioned object attribute recognition method. The memory 1104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 1104 may further include memory remotely located relative to the processor 1102, and these remote memories can be connected to the object attribute recognition device 110 via a network. Examples of the above-mentioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0207] The transmission device 1106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the object attribute identification device 110. In one example, the transmission device 1106 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In one embodiment, the transmission device 1106 may be a radio frequency (RF) module for wireless communication with the Internet.

[0208] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows a user to interact with the user interface of the object attribute recognition device 110 (or mobile device).

[0209] Embodiments of this application also provide a computer-readable storage medium, which can be disposed in an object attribute recognition device to store at least one instruction or at least one program related to implementing the object attribute recognition method in the method embodiment. The at least one instruction or the at least one program is loaded and executed by the processor to implement the object attribute recognition method provided in the above method embodiment.

[0210] Optionally, in this embodiment, the storage medium may be located in at least one of the multiple network servers in a computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0211] Embodiments of this application also provide a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the object attribute identification method as provided in the method embodiments.

[0212] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired results. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are also possible or may be advantageous.

[0213] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and apparatus embodiments are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0214] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0215] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0216] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method of object attribute recognition, characterized by, The method comprises: obtaining identification text feature information of a to-be-identified object and a plurality of interaction record texts corresponding to the to-be-identified object, each interaction record text being associated text generated by an interaction object in the process of resource interaction with the to-be-identified object; extracting interaction keywords from the plurality of interaction record texts to obtain a plurality of interaction keywords; inputting the plurality of interaction keywords into a keyword feature extraction model to extract keyword features, and obtaining interaction feature information; based on the identification text feature information and the interaction feature information, identifying the to-be-identified object as a merchant attribute to obtain merchant attribute indication information, the merchant attribute indication information being used to indicate whether the to-be-identified object has a merchant attribute.

2. The method of claim 1, wherein, The plurality of interaction record texts are generated in a plurality of sub-time periods, and the extracting of the interaction keywords from the plurality of interaction record texts comprises: performing word segmentation processing on at least one target text generated in a target sub-time period to obtain a plurality of text segments, the target sub-time period being any one of the plurality of sub-time periods, and the at least one target text being at least one interaction record text generated in the target sub-time period from the plurality of interaction record texts; determining a word proportion index of each text segment in the at least one target text; determining a time period proportion index of a covered sub-time period corresponding to each text segment in the plurality of sub-time periods, the covered sub-time period corresponding to each text segment being a sub-time period in which the at least one target text is generated and contains the corresponding text segment; based on the word proportion index and the time period proportion index, performing text importance analysis on each text segment with respect to the target sub-time period to obtain a first importance representation index of each text segment; based on the first importance representation index of each of the plurality of text segments, determining at least one initial keyword corresponding to the target sub-time period from the plurality of text segments; using the at least one initial keyword corresponding to each of the plurality of sub-time periods as the plurality of interaction keywords.

3. The method of claim 1, wherein, The extracting of the interaction keywords from the plurality of interaction record texts comprises: performing word segmentation processing on the plurality of interaction record texts to obtain a plurality of candidate phrases and at least one segment included in each candidate phrase; performing context influence analysis on the at least one segment included in each candidate phrase to obtain a second importance representation index of each candidate phrase; based on the second importance representation index of each of the plurality of candidate phrases, determining the plurality of interaction keywords from the plurality of candidate phrases.

4. The method of claim 3, wherein, The context influence analysis on the at least one segment included in each candidate phrase to obtain the second importance representation index of each candidate phrase comprises: determining the number of co-occurrences of each segment included in each candidate phrase in the plurality of candidate phrases; determining the number of occurrences of each segment in the plurality of candidate phrases; determine a context influence indicator of each of the segmented words based on a ratio between the co-occurrence number of each of the segmented words and the occurrence number of each of the segmented words; perform indicator fusion on the context influence indicators of at least one segmented word included in each of the candidate phrases to obtain a second importance representation indicator of each of the candidate phrases.

5. The method of claim 1, wherein, The generation time period of the plurality of interaction record texts includes a plurality of sub-time periods, and the plurality of interaction keywords include at least one initial keyword corresponding to each of the plurality of sub-time periods. The keyword feature extraction model includes a segmented word semantic extraction layer, a time period text semantic extraction layer, a time sequence feature extraction layer, and a feature aggregation layer. The keyword feature extraction includes: performing splicing processing on the at least one initial keyword corresponding to each of the plurality of sub-time periods in a time sequence of the plurality of sub-time periods to obtain a target spliced text; inputting the target spliced text into the segmented word semantic extraction layer to perform segmented word semantic extraction on the at least one initial keyword corresponding to each of the plurality of sub-time periods to obtain segmented word semantic feature information corresponding to each initial keyword; inputting the target spliced text into the time period text semantic extraction layer to perform semantic extraction on the at least one initial keyword corresponding to the sub-time period to which each initial keyword belongs to obtain time period text feature information corresponding to each initial keyword; inputting the target spliced text into the time sequence feature extraction layer to perform time sequence feature extraction on the sub-time period to which each initial keyword belongs to obtain time sequence feature information corresponding to each initial keyword; inputting the segmented word semantic feature information, the time period text feature information, and the time sequence feature information into the feature aggregation layer to perform context semantic fusion to obtain the interaction feature information.

6. The method according to any one of claims 1 to 5, characterized in that, The interaction keyword extraction includes: performing interaction keyword extraction on the plurality of interaction record texts to obtain a plurality of initial keywords; performing part-of-speech filtering on the plurality of initial keywords to obtain the plurality of interaction keywords.

7. The method according to any one of claims 1 to 5, characterized in that, The merchant attribute identification includes: inputting the identification text feature information and the interaction feature information into a feature fusion model to perform feature fusion processing to obtain target attribute feature information; inputting the target attribute feature information into a merchant attribute identification model to perform attribute identification to obtain the merchant attribute indication information.

8. An object attribute recognition apparatus characterized by comprising: The apparatus includes: a data acquisition module configured to acquire identification text feature information of a to-be-identified object and a plurality of interaction record texts corresponding to the to-be-identified object, each interaction record text being associated text generated by an interaction object in a process of resource interaction with the to-be-identified object; an interaction keyword extraction module configured to perform interaction keyword extraction on the plurality of interaction record texts to obtain a plurality of interaction keywords; and an attribute identification module configured to perform merchant attribute identification on the to-be-identified object based on the identification text feature information and the interaction feature information to obtain merchant attribute indication information. The keyword feature extraction module is configured to input the plurality of interaction keywords into a keyword feature extraction model to perform keyword feature extraction and obtain interaction feature information. The merchant attribute identification module is configured to perform merchant attribute identification on the to-be-identified object based on the identified text feature information and the interaction feature information, and obtain merchant attribute indication information, which is used to indicate whether the to-be-identified object has a merchant attribute.

9. An object attribute identifying apparatus characterized by comprising: The device comprises a processor and a memory, and the memory stores at least one instruction or at least one program, which is loaded and executed by the processor to implement the object attribute identification method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction or at least one program, which is loaded and executed by the processor to implement the object attribute identification method according to any one of claims 1 to 7.

11. A computer program product, characterised in that, The computer program product comprises at least one instruction or at least one program, which is loaded and executed by the processor to implement the object attribute identification method according to any one of claims 1 to 7.

Images