Information processing method and system

Abstract

Embodiments of the present specification provide an information processing method and system, wherein the information processing method comprises: obtaining object query information; inputting the object query information into a semantic identifier generation model for processing to obtain query semantic identifier information corresponding to the object query information, and determining object semantic identifier information corresponding to candidate objects in a candidate object set respectively; determining an identifier information matching value between the query semantic identifier information and the object semantic identifier information according to pre-set hierarchical matching information; and filtering a target object matching the object query information from the candidate object set based on the identifier information matching value.

Description

Technical Field

[0001] The embodiments in this specification relate to the field of information processing technology, and in particular to information processing methods and systems. Background Technology

[0002] With the rapid development of computer and internet technologies, users have placed higher demands on the accuracy and relevance of search results. In the core process of relevance calculation in search, traditional methods primarily rely on multimodal or textual embedding vectors, measuring the degree of match between the query and the target object by calculating the similarity between vectors. However, while continuous embedding vectors can capture macroscopic semantic features, they have significant limitations: their high-dimensional continuous space contains a large amount of redundant noise, making it difficult to effectively distinguish fine-grained semantic differences (such as the style, material, and applicable scenarios of related products in e-commerce search scenarios), especially in professional domain queries. Furthermore, existing models often lack the ability to dynamically understand user query intent, leading to discrepancies between search results and actual needs. Therefore, an effective solution is urgently needed to address these problems. Summary of the Invention

[0003] In view of this, embodiments of this specification provide information processing methods. One or more embodiments of this specification also relate to information processing systems, information processing apparatuses, computing devices, computer-readable storage media, and computer program products, to address technical deficiencies in the prior art.

[0004] According to a first aspect of the embodiments of this specification, an information processing method is provided, comprising: Retrieve object query information; The object query information is input into the semantic identifier generation model for processing to obtain the query semantic identifier information corresponding to the object query information, and to determine the object semantic identifier information corresponding to each candidate object in the candidate object set. Based on preset hierarchical matching information, determine the matching value of the identifier information between the query semantic identifier information and the object semantic identifier information; Based on the identification information matching value, target objects that match the object query information are filtered from the candidate object set.

[0005] According to a second aspect of the embodiments of this specification, another information processing method is provided, applied to a transaction server, including: Receive product query information submitted by the client through the product query page; The product query information is input into the semantic identifier generation model for processing to obtain the query semantic identifier information corresponding to the product query information, and to determine the product semantic identifier information corresponding to each candidate product in the candidate product set. Based on preset hierarchical matching information, determine the matching value of the identifier information between the query semantic identifier information and the product semantic identifier information; Based on the identification information matching value, target products that match the product query information are selected from the candidate product set, and the product information corresponding to the target products is fed back to the client.

[0006] According to a third aspect of the embodiments of this specification, an information processing system is provided, including a client and a server, comprising: The client is configured to determine object query information in response to a query request submitted through the object query page, and send the object query information to the server. The server is configured to input the object query information into a semantic identifier generation model for processing, obtain the query semantic identifier information corresponding to the object query information, and determine the object semantic identifier information corresponding to each candidate object in the candidate object set; determine the identifier information matching value between the query semantic identifier information and the object semantic identifier information according to preset hierarchical matching information; based on the identifier information matching value, filter the target objects that match the object query information in the candidate object set, and feed back the object information corresponding to the target objects to the client.

[0007] According to a fourth aspect of the embodiments of this specification, an information processing apparatus is provided, comprising: The retrieval module is configured to retrieve object query information; The processing module is configured to input the object query information into the semantic identifier generation model for processing, obtain the query semantic identifier information corresponding to the object query information, and determine the object semantic identifier information corresponding to each candidate object in the candidate object set. The determination module is configured to determine the identification information matching value between the query semantic identification information and the object semantic identification information according to preset hierarchical matching information; The filtering module is configured to filter target objects that match the object query information from the candidate object set based on the identification information matching value.

[0008] According to a fifth aspect of the embodiments of this specification, another information processing apparatus is provided, applied to a transaction server, comprising: The receiving module is configured to receive product query information submitted by the client through the product query page; The processing module is configured to input the product query information into the semantic identifier generation model for processing, obtain the query semantic identifier information corresponding to the product query information, and determine the product semantic identifier information corresponding to each candidate product in the candidate product set; The determination module is configured to determine the identification information matching value between the query semantic identification information and the product semantic identification information according to preset hierarchical matching information; The feedback module is configured to filter target products that match the product query information from the candidate product set based on the identification information matching value, and to feed back the product information corresponding to the target product to the client.

[0009] According to a sixth aspect of the embodiments of this specification, a computing device is provided, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the above-described information processing method.

[0010] According to a seventh aspect of the embodiments of this specification, a computer-readable storage medium is provided that stores computer-executable instructions, which, when executed by a processor, implement the steps of the information processing method described above.

[0011] According to an eighth aspect of the embodiments of this specification, a computer program product is provided, including a computer program or instructions that, when executed by a processor, implement the steps of the information processing method described above.

[0012] The information processing method provided in this embodiment, in order to provide more accurate relevance calculation through discretized semantic identifier information, first inputs the object query information into a semantic identifier generation model after obtaining the object query information. This enables the semantic identifier generation model to construct query semantic identifier information corresponding to the object query information, and simultaneously determines the object semantic identifier information corresponding to each candidate object in the candidate object set. After obtaining the semantic identifier information corresponding to the query information and the candidate objects respectively, the identifier information matching value between the query semantic identifier information and the object semantic identifier information can be determined according to the preset hierarchical matching information. The identifier information matching value can reflect the degree of association between the query information and each candidate object. Finally, based on the identifier information matching value, the target objects that match the object query information can be filtered from the candidate object set. This method improves the accuracy of relevance calculation between query information and candidate objects in object query scenarios by introducing semantic identifier information. Furthermore, the use of hierarchical matching information to determine the final value further ensures that the value accurately reflects the true relevance between the query information and the candidate objects. The selected target objects can accurately match the object query information, thus facilitating downstream service use. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating an information processing method provided in one embodiment of this specification; Figure 2a This is a schematic diagram of information processing in an information processing method provided in one embodiment of this specification; Figure 2b This is a schematic diagram of information processing in another information processing method provided in one embodiment of this specification; Figure 3 This is a flowchart of another information processing method provided in one embodiment of this specification; Figure 4 This is a flowchart illustrating the processing procedure of an information processing method provided in one embodiment of this specification. Figure 5 This is a schematic diagram of the structure of an information processing system provided in one embodiment of this specification; Figure 6 This is a schematic diagram of the structure of an information processing device provided in one embodiment of this specification; Figure 7 This is a schematic diagram of the structure of another information processing device provided in one embodiment of this specification; Figure 8 This is a structural block diagram of a computing device provided in one embodiment of this specification. Detailed Implementation

[0014] Many specific details are set forth in the following description to provide a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar extensions without departing from the spirit of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0015] The terminology used in one or more embodiments of this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the one or more embodiments of this specification. The singular forms “a,” “described,” and “the” as used in one or more embodiments of this specification and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0016] It should be understood that although the terms first, second, etc., may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first may also be referred to as second without departing from the scope of one or more embodiments of this specification, and similarly, second may also be referred to as first. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0017] Furthermore, it should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0018] The technical solutions provided in this application can employ deep learning models with relatively large parameter scales. However, this large model is merely an example; this application does not limit the number of model parameters supported by the deep learning model used, aiming to meet actual needs. The deep learning models involved in this application can be artificial intelligence-based language models (LM) or multimodal models (MM). First, the terms and concepts used in one or more embodiments of this specification will be explained.

[0019] SID (Semantic Identifier) ​​is a discrete semantic representation technique that enables fine-grained explicit semantic modeling by encoding continuous semantic information into interpretable symbol sequences (such as IDs or tags). Its advantages include robustness to noise, strong interpretability, and support for semantic combination expressions (such as combinations of product attributes), making it suitable for long-tail query matching and multi-attribute retrieval tasks in e-commerce search.

[0020] Supervised Fine-tuning (SFT) is an optimization method for pre-trained models. By continuing training on labeled data for specific tasks, the model can be adapted to the needs of e-commerce scenarios (such as product ranking and intent recognition). Its core value lies in the balance between task adaptability, controllability, and efficiency. Performance can be improved by updating only a few parameters, making it a key technology connecting general-purpose models with vertical domain applications.

[0021] RQ-VAE (Residual Quantized Variational AutoEncoder) is a deep discretization model that compresses high-dimensional data (such as product images) into low-dimensional discrete codebooks through hierarchical quantization and residual connections. Its hierarchical structure captures semantics at different granularities, residual learning reduces information loss, and it supports end-to-end training.

[0022] MT5 (Multilingual Text-to-Text Transfer Transformer) is a multilingual pre-trained model developed based on the T5 architecture, supporting text generation and understanding in 100+ languages. Its unified text-to-text framework simplifies task design, and cross-language parameter sharing enables knowledge transfer.

[0023] To address the aforementioned technical problems, this specification provides an information processing method. One or more embodiments of this specification also relate to an information processing system, an information processing apparatus, a computing device, a computer-readable storage medium, and a computer program product, which will be described in detail in the following embodiments.

[0024] In practical applications, discrete representation techniques (such as Symbolic ID, SID) have demonstrated unique advantages in fields such as natural language processing and recommender systems. By discretizing semantic information into interpretable units through symbolic encoding, they can more accurately capture key features and reduce noise interference. Meanwhile, generative paradigms, with their powerful contextual modeling capabilities, offer new insights into dynamically understanding user intent. However, in e-commerce search scenarios, no research has systematically explored the combined application of SID and generative paradigms, and existing models still lack sufficient discriminative power for fine-grained product attributes. Therefore, an effective solution is urgently needed to address these issues.

[0025] The information processing method provided in this embodiment, in order to provide more accurate relevance calculation through discretized semantic identifier information, first inputs the object query information into a semantic identifier generation model after obtaining the object query information. This enables the semantic identifier generation model to construct query semantic identifier information corresponding to the object query information, and simultaneously determines the object semantic identifier information corresponding to each candidate object in the candidate object set. After obtaining the semantic identifier information corresponding to the query information and the candidate objects respectively, the identifier information matching value between the query semantic identifier information and the object semantic identifier information can be determined according to the preset hierarchical matching information. The identifier information matching value can reflect the degree of association between the query information and each candidate object. Finally, based on the identifier information matching value, the target objects that match the object query information can be filtered from the candidate object set. This method improves the accuracy of relevance calculation between query information and candidate objects in object query scenarios by introducing semantic identifier information. Furthermore, the use of hierarchical matching information to determine the final value further ensures that the value accurately reflects the true relevance between the query information and the candidate objects. The selected target objects can accurately match the object query information, thus facilitating downstream service use.

[0026] See Figure 1 , Figure 1 A flowchart of an information processing method according to an embodiment of this specification is shown, which specifically includes the following steps.

[0027] Step S102: Obtain object query information.

[0028] The information processing method provided in this embodiment can be applied to any search scenario. For example, in an e-commerce search scenario, it can search for and recommend products that the user wants to buy within the shopping platform based on the query information submitted by the user, or it can search for and recommend products that the user is interested in within the shopping platform based on the set query information. In another example, in a content search scenario, it can search for and recommend content such as images, text, and videos (e.g., travel guides, shopping guides, brand clothing sharing content) that the user wants to browse within a content sharing platform based on the query information submitted by the user. In yet another example, in a multimedia resource search scenario, it can search for and recommend resources such as audio, video, text, and images (e.g., movies, comics, music, papers) that the user wants to browse within a multimedia resource browsing platform based on the query information submitted by the user. In any scenario with a query requirement, the information processing method provided in this embodiment can complete a fast and accurate query operation to provide high-quality query results to downstream users, thereby meeting different service needs.

[0029] This embodiment uses an e-commerce search scenario as an example to illustrate the information processing method. The same or corresponding descriptions in other scenarios can be found in the description of this embodiment, and will not be elaborated on further here.

[0030] Specifically, object query information refers to the query information submitted by upstream nodes (such as users) in a search scenario. This information can be in text, image, audio, or video format, or a combination of these formats. The query information is user-defined, and this embodiment does not impose any limitations. For example, in a scenario where a user is searching for products on a shopping platform, the query information submitted could be the query text "{red coat}", or images and descriptions of similar products. This allows for the subsequent construction of semantic identifiers based on the query information, enabling product search operations from a semantic identifier perspective, reducing computational complexity while improving query efficiency.

[0031] Based on this, in order to provide more accurate relevance calculations through discretized semantic identifier information, after obtaining the object query information, the object query information can be first input into the semantic identifier generation model. This allows the semantic identifier generation model to construct the query semantic identifier information corresponding to the object query information, and simultaneously determine the object semantic identifier information corresponding to each candidate object in the candidate object set. After obtaining the semantic identifier information corresponding to the query information and the candidate objects, the identifier information matching value between the query semantic identifier information and the object semantic identifier information can be determined according to the preset hierarchical matching information. The identifier information matching value can reflect the degree of association between the query information and each candidate object. Finally, based on the identifier information matching value, the target object matching the object query information can be selected from the candidate object set.

[0032] In other words, the information processing method provided in this embodiment can improve the accuracy of the correlation calculation between query information and candidate objects by introducing semantic identification information in object query scenarios. Furthermore, it uses hierarchical matching information to complete the final value determination, which can further ensure that the identification information matching value can accurately reflect the true correlation between query information and candidate objects. The target objects selected in this way can accurately hit the object query information, thereby facilitating the use of downstream services.

[0033] Step S104: Input the object query information into the semantic identifier generation model for processing to obtain the query semantic identifier information corresponding to the object query information, and determine the object semantic identifier information corresponding to each candidate object in the candidate object set.

[0034] Specifically, the semantic identifier generation model refers to a language model that can construct corresponding semantic identifier information based on the input object query information or the object information corresponding to the candidate objects. This model can incorporate pre-defined prompt words during the construction of semantic identifier information, thereby ensuring more accurate output results and meeting downstream usage requirements. The query semantic identifier information specifically refers to the semantic ID corresponding to the object query information. Correspondingly, the candidate object set refers to the set of objects that need to be filtered to match the object query information; these can be products, content, videos, audio, etc., without any limitation in this embodiment. The object semantic identifier information is the semantic ID corresponding to each candidate object in the candidate object set. This ID can be pre-constructed and stored in a database for reuse during the application phase, or it can be generated in real time, without any limitation in this embodiment.

[0035] It should be noted that semantic identifier information contains multiple sub-identifier information, and each sub-identifier information represents a different level of type, with a gradient relationship between different levels of type. For example, if the object query information is {A brand computer case}, the semantic ID generated is {1,101,186}, where 1 represents the electronic product type, 101 represents the computer case type under the electronic product type, and 186 can represent brand A. Subsequent query operations using semantic IDs with this structure can ensure that the filtered objects are more closely matched to the object query information, thereby facilitating downstream service use.

[0036] Based on this, after obtaining the object query information as described above, in order to achieve fast and accurate object filtering and determine the target object matching the object query information, semantic identification information corresponding to the candidate objects and the object query information can be constructed in the early preparation stage. This allows the relevance calculation between the candidate objects and the object query information to be completed through the semantic identification information.

[0037] It's important to note that calculating relevance based on semantic identifiers differs from calculating relevance based on feature representations. Feature-based relevance calculations typically use continuous vectors (such as embeddings) to represent objects and queries, measuring relevance by calculating cosine similarity, Euclidean distance, etc. This calculation relies heavily on global similarity and may overlook key local features. For example, the vectors for the query "red dress" and the product "blue dress" might score highly due to similarity in other dimensions, even if their actual color attributes don't match. In contrast, calculating relevance between objects and queries based on semantic identifiers is more accurate in fine-grained feature matching and combination. For instance, the query's ID sequence needs to perfectly match the product's ID sequence in key attributes (such as color and size), or the relevance can be dynamically adjusted using a rule engine (such as attribute weights).

[0038] Furthermore, relevance calculations based on feature representations are often performed using black-box algorithms, making it difficult to explain why certain objects rank highly. In contrast, relevance calculations based on semantic identifiers can clearly define the semantics, resulting in more interpretable results. Therefore, it can be determined that using semantic identifiers to complete subsequent target object filtering not only has stronger noise resistance but also improves fine-grained matching accuracy (continuous vectors may introduce noise due to data bias or model limitations (such as making the vectors for "sports shoes" and "casual shoes" too close), while semantic IDs avoid noise interference by explicitly encoding key attributes. For example, querying the ID sequence of "A running shoes" can accurately match the brand, type, and purpose in the product attributes, reducing false recalls). It can also support the combination of semantics and complex logic (semantic IDs can represent multiple attribute combinations (such as "summer + cotton + short sleeves"), while continuous vectors need to capture the combination relationship through complex models (such as attention mechanisms). In e-commerce scenarios, user queries often contain multiple constraints (such as "white, XL size, free shipping"), and semantic IDs can directly satisfy all conditions through symbol matching, resulting in higher computational efficiency), thus meeting the needs of fast and accurate filtering operations in different scenarios and satisfying the downstream usage requirements of different service scenarios.

[0039] Furthermore, when constructing semantic identifier information using a semantic identifier generation model, the operation can be completed through the model's internal encoding module and codebook quantization module. In this embodiment, the step of inputting the object query information into the semantic identifier generation model for processing to obtain the query semantic identifier information corresponding to the object query information includes: The object query information is input into a semantic identifier generation model, wherein the semantic identifier generation model includes an encoding module and a codebook quantization module; the encoding module is used to encode the object query information to obtain a query encoding vector corresponding to the object query information; the codebook quantization module is used to perform multi-level residual quantization on the query encoding vector, and the query semantic identifier information corresponding to the object query information is determined based on the processing result.

[0040] Specifically, the encoding module refers to the module in the semantic identifier generation model used to encode the input content, constructing the encoding vector corresponding to the input content. Correspondingly, the codebook quantization module refers to the module that performs multi-level residual quantization on the encoded vector, constructing the semantic ID corresponding to the input content. Here, the query encoding vector is the encoding vector corresponding to the object query information, and multi-level residual quantization specifically refers to the operation of calculating codewords and residuals at different levels on the encoding vector, with the aim of mapping the encoding vector to the true expression of the semantic ID.

[0041] Based on this, when constructing the query semantic identifier information corresponding to the object query information, the object query information can first be input into the semantic identifier generation model, which includes an encoding module and a codebook quantization module. On this basis, the encoding module can be used to encode the object query information to obtain the query encoding vector corresponding to the object query information. Then, the codebook quantization module is used to perform multi-level residual quantization on the query encoding vector, so that the query semantic identifier information corresponding to the object query information can be determined according to the processing result, so as to be used for subsequent relevance calculation.

[0042] In practical applications, the mT5 encoder can be combined with RQ-VAE multi-layer residual quantization to generate semantic IDs corresponding to object query information. During the semantic encoding stage, the mT5 encoder architecture can be based on the multilingual Transformer's Encoder module, consisting of 12 stacked Transformer layers. Each layer captures the semantic context of the text through a multi-head self-attention mechanism. The input text is mapped to a 768-dimensional vector via word embedding layers, and after bidirectional encoding, a continuous semantic vector is output. For example, the encoding result of "smartphone" will activate semantic dimensions related to "portable device" and "communication function." Vector optimization can then be performed, where average pooling can be used to compress the sequence dimension, reducing the hidden state matrix to a single 768-dimensional vector. For long texts, dynamic segmented pooling can also be used to preserve local semantic features and avoid information loss.

[0043] Furthermore, in the residual quantization stage, the RQ-VAE core structure can output a continuous vector z from the encoding module, which can then enter a three-level residual quantization module. The hierarchical quantization mechanism is as follows: the first-level codebook (size 1024) performs coarse-grained quantization, outputting codeword c0 and residual r1 = z - c0; the second level quantizes r1 twice to obtain c1, and the residual r2 = r1 - c1; the final level processes r2 to generate c2. The final semantic ID is represented as a triple (c0, c1, c2), theoretically representing 1024³ semantic combinations. Based on this, the codebook is dynamically updated: an exponential moving average (EMA) strategy can be used to update the codebook vector. In the initial stage, the codebook is initialized using K-means clustering of the first batch of data vectors to ensure that the codebook covers the real data distribution. In addition, during training, the codebook learning rate can be set to 10 times that of the encoder to accelerate convergence.

[0044] Furthermore, after residual quantization, during semantic ID generation, the level 3 codewords are mapped to independent indexes via a lookup table, and then concatenated to form a unique semantic ID. For example, "high-end mobile phone" might be mapped as (Technology Category_512, Brand_A_34, Positioning_Flagship_99). This ID supports similarity calculation; for example, the Hamming distance between the IDs of the A brand mobile phone model 15 and model 15 Pro can be less than 2, reflecting semantic relevance. This ensures that the final semantic ID is more accurate for subsequent relevance calculations.

[0045] In summary, the encoding module transforms object query information into high-dimensional query encoding vectors, effectively capturing multimodal semantic features; the codebook quantization module adopts a multi-layer residual quantization mechanism, which maps continuous vectors to discrete semantic identifiers while maintaining semantic integrity, realizing structured compression of query information, improving retrieval efficiency and cross-modal matching accuracy, and reducing storage and computational overhead.

[0046] Furthermore, when determining the semantic identifier information corresponding to each candidate object in the candidate object set, it is considered that there may be newly added candidate objects in the candidate object set, while other historical candidate objects already have corresponding semantic identifier information due to calculation. Newly added candidate objects can be temporarily generated, and the generated semantic identifier information can be stored for use in subsequent query operations. In this embodiment, determining the semantic identifier information corresponding to each candidate object in the candidate object set includes: Historical candidate objects and newly added candidate objects are determined from the candidate object set, and the semantic identification information of the objects corresponding to the historical candidate objects is extracted from the identification information repository; the object title information corresponding to the newly added candidate objects is determined, and the object title information is input into the semantic identification generation model for processing to obtain the object semantic identification information corresponding to the newly added candidate objects.

[0047] Specifically, historical candidate objects refer to candidate objects in the candidate object set whose corresponding semantic identifier information already exists. New candidate objects refer to candidate objects whose semantic identifier information has not yet been constructed. Correspondingly, the identifier information repository refers to a database that stores the semantic identifier information corresponding to candidate objects, facilitating the reuse of this information during the application phase and saving time in generating semantic identifier information. Correspondingly, object title information refers to the title text information corresponding to new candidate objects. For example, if a new candidate object is a newly listed product, the product's title can be used as its corresponding title information to construct a semantic ID.

[0048] Based on this, in order to ensure that the subsequent relevance calculation can cover all candidate objects within the platform, historical candidate objects and new candidate objects can be identified from the candidate object set. For historical candidate objects, the semantic identifier information corresponding to the historical candidate objects can be directly extracted from the identifier information repository for reuse. For new candidate objects, since the semantic identifier information has not yet been constructed, the object title information corresponding to the new candidate objects can be determined. At this time, the object title information can be input into the semantic identifier generation model for processing, thereby obtaining the object semantic identifier information corresponding to the new candidate objects. After determining the object semantic identifier information corresponding to each candidate object, the subsequent relevance calculation operation can be performed.

[0049] Furthermore, since the semantic identifier information corresponding to newly added candidate objects is generated temporarily during the application phase, it can be stored in an identifier information repository for reuse during subsequent relevance calculations, effectively saving object query time. Writing object semantic identifier information to the identifier information repository can be done in key-value pair format for easier management.

[0050] For example, on shopping platform A, a user enters the product query "{A brand computer case}" into the search box of the shopping app. To display the corresponding product on the search results page, the query "{A brand computer case}" can first be input into a semantic identifier generation model for processing. Based on the processing result, the semantic ID corresponding to the query "{A brand computer case}" can be determined as {1, 101, 186}. Simultaneously, to achieve product matching based on semantic IDs, it is also necessary to determine the product semantic IDs of candidate products on shopping platform A. For candidate products in the candidate product set without constructed semantic IDs, the semantic identifier generation model can also process the title information of the candidate products to generate their corresponding product semantic IDs. This allows for subsequent filtering of target products by combining the product query information and the semantic IDs of the candidate products.

[0051] In summary, by using an identifier information repository to store the semantic identifier information of historical candidate objects, the semantic identifier information can be directly reused in object query scenarios, thereby saving computation time and accelerating object query efficiency.

[0052] Furthermore, to enable the semantic tag generation model to generate semantic tag information corresponding to both query information and candidate products, a two-stage training model can be used to complete the model training operation. In this embodiment, the training of the semantic tag generation model includes: Obtain query object sample pairs, wherein the query object sample pairs include sample query information and sample objects; use the query object sample pairs to perform a first-stage training on an initial semantic identifier generation model to obtain an intermediate semantic identifier generation model; use the query object sample pairs to perform a second-stage training on the intermediate semantic identifier generation model to obtain the semantic identifier generation model; wherein the first-stage training is used for the semantic identifier generation model to learn the construction of object semantic identifier information, and the second-stage training is used for the semantic identifier generation model to learn the construction of query semantic identifier information.

[0053] Specifically, a query object sample pair refers to a sample data pair composed of sample query information and sample objects related to the query information. The sample objects related to the query information can be positively or negatively related, used to form positive and negative sample pairs for subsequent model training. Correspondingly, the initial semantic label generation model refers to the semantic label generation model to be trained. The intermediate semantic label generation model refers to the semantic label generation model obtained after completing the first stage of training. The first stage of training specifically refers to the training period used for the semantic label generation model to learn the ability to construct semantic label information corresponding to objects. This can be understood as the training period for the semantic label generation model to learn the ability to generate semantic label information corresponding to objects. The second stage of training specifically refers to the training period used for the semantic label generation model to learn the construction of semantic label information corresponding to queries. This can be understood as the training period for the semantic label generation model to learn the ability to generate semantic label information corresponding to query information.

[0054] Based on this, during the model training phase, query object sample pairs, including sample query information and sample objects, can be obtained first. Then, the initial semantic label generation model can be trained in the first stage using the query object sample pairs to enable the model to learn the ability to construct semantic label information of objects, thereby obtaining an intermediate semantic label generation model. After completing the first stage of training, the intermediate semantic label generation model can be trained in the second stage using the query object sample pairs to enable the model to learn the ability to construct semantic label information of queries. After completing the first and second stages of training, the semantic label generation model that meets the training stopping condition can be deployed to the service scenario for use.

[0055] It should be noted that the first stage of training must be completed before the second stage. The purpose is to use the samples from the first stage as supervision signals for the second stage, thereby improving the training effect. After both stages of training, the model can be validated using samples from the validation set, ensuring that the model deployed in the service scenario provides stable and accurate predictive capabilities.

[0056] In summary, by training the semantic tag generation model in two stages, and by having the model learn the tag generation capabilities for different types of information at different stages, the model can provide stable and accurate services after deployment, thereby supporting downstream services to complete query operations based on tag information and improving query accuracy.

[0057] Based on this, during the first stage of training, the quantized representations of relevant samples can be made closer in the semantic space, while the representations of irrelevant samples can be made further apart, thereby enabling the model to learn the ability to generate semantic identifier information about objects. In this embodiment, the first stage of training the initial semantic identifier generation model using the query object samples to obtain an intermediate semantic identifier generation model includes: The sample query information and the sample object are input into an initial semantic identifier generation model, which includes an initial encoding module and an initial codebook quantization module. The initial encoding module encodes the sample query information and the sample object respectively to obtain a sample query encoding vector corresponding to the sample query information and a sample object encoding vector corresponding to the sample object. The initial codebook quantization module performs multi-level residual quantization on the sample query encoding vector and the sample object encoding vector to obtain a discrete semantic identifier information sequence. Based on the first sample label corresponding to the query object sample and the discrete semantic identifier information sequence, the initial semantic identifier generation model is trained through comparative learning to obtain an intermediate semantic identifier generation model.

[0058] Specifically, the initial encoding module and the initial codebook quantization module refer to the encoding module and codebook quantization module before parameter tuning. Correspondingly, the sample query encoding vector is the encoded feature obtained after encoding the sample query information. The discrete semantic identifier information sequence specifically refers to the sequence of discrete semantic identifier information generated by the initial codebook quantization module after performing multi-level residual quantization on the sample query encoding vector and the sample object encoding vector. This sequence is used for model parameter tuning after comparison with the label. The first sample label is the label corresponding to the sample.

[0059] Based on this, in the first stage of training, the sample query information and sample objects can be input into the initial semantic label generation model. At this time, the initial encoding module in the initial semantic label generation model can be used to encode the sample query information and sample objects respectively to obtain the sample query encoding vector corresponding to the sample query information and the sample object encoding vector corresponding to the sample object. On this basis, the initial codebook quantization module can be used to perform multi-layer residual quantization on the sample query encoding vector and the sample object encoding vector to construct a discrete semantic label information sequence based on the processing results. After that, the initial semantic label generation model can be trained by comparison with the first sample label and the discrete semantic label information sequence corresponding to the query object sample to obtain the intermediate semantic label generation model, and then the subsequent second stage of training can be carried out.

[0060] For specific implementation, please refer to Figure 2a The diagram illustrates how, to ensure the model possesses strong predictive power and better robustness, training samples can be constructed using real-world interaction data. For example, in e-commerce scenarios, user click and transaction data within a defined time period can be collected to form a set number of query-product pairs, serving as a positive sample set for relevance training. The first stage of training aims to enable the model to generate semantic IDs corresponding to products. During this process, a multilingual text-to-text transfer Transformer (mT5) encoder can be used to semantically encode the input query text and product text (such as product title information), resulting in the text representation vector Q.Emb for the query text and the text representation vector I.Emb for the product. Then, both encoded vectors can be fed into the RQ-VAE for training, realizing the input of text representation vectors into the codebook quantization module of the Residual Quantization Variational Autoencoder (RQ-VAE). Through a multi-layer residual quantization process, continuous vectors are mapped into discrete semantic ID sequences. For related query-product pairs, the Information Noise Contrast Estimation (InfoNCE) loss function is used for contrastive learning within the same batch, making the quantized representations of related samples closer in semantic space and the representations of unrelated samples further apart. After complete training, the model can learn to construct a system for the semantic ID representation of products.

[0061] In summary, the first stage of training enables the semantic tag generation model to learn the ability to construct semantic tag information about objects, thereby making the model more accurate in constructing semantic tag information corresponding to candidate objects, so as to provide a faster and more accurate basis for relevance calculation when applied.

[0062] In the second stage of training, a generative training paradigm can be used to enable the model to learn the ability to construct query semantic identifier information. In this embodiment, the second stage training of the intermediate semantic identifier generation model using the query object samples to obtain the semantic identifier generation model includes: Obtain the semantic identifier information of the sample objects associated with the first stage training, and use the semantic identifier information of the sample objects as the second sample label; input the sample query information into the intermediate semantic identifier generation model for processing to obtain the predicted object semantic identifier information sequence corresponding to the sample query information; perform generative training on the intermediate semantic identifier generation model based on the second sample label and the predicted object semantic identifier information sequence to obtain the semantic identifier generation model.

[0063] Specifically, the semantic identifier information of the sample objects refers to the semantic identifier information of the sample objects during the first stage of training, which is used as the second sample label for parameter tuning during model training. Correspondingly, the predicted object semantic identifier information sequence refers to the sequence of semantic identifier information corresponding to the sample objects output by the model. Correspondingly, generative training refers to training that optimizes the model's sequence generation ability by using the semantic identifier information corresponding to the sample objects as the target output.

[0064] Based on this, during the second stage of training, the semantic identifier information of the sample objects associated with the first stage of training can be obtained. At this time, the semantic identifier information of the sample objects can be used as the second sample label. Then, the sample query information can be input into the intermediate semantic identifier generation model for processing to obtain the sequence of predicted object semantic identifier information corresponding to the sample query information. After that, the intermediate semantic identifier generation model is generatively trained by combining the second sample label and the sequence of predicted object semantic identifier information to obtain the semantic identifier generation model.

[0065] In other words, see Figure 2b As shown in the diagram, during the second stage of training, the product semantic IDs obtained in the first stage can be used as supervision signals. The query text is input into the intermediate semantic identifier generation model, and the training model directly generates a sequence of product semantic IDs related to the query. By adopting a generative training paradigm and using product semantic IDs as the target output, the sequence generation capability of the model is optimized. After complete training, the semantic identifier generation model can be equipped with the ability to generate query semantic identifiers. After deployment, it can provide a stable semantic ID construction capability.

[0066] In summary, by enabling the model to learn the ability to construct semantic identifiers for queries during the second stage of training, the model can simultaneously possess the ability to construct semantic identifiers for both objects and queries. This allows for faster identifier construction in subsequent computational processing, thereby saving computational resources and improving processing efficiency.

[0067] Step S106: Determine the identification information matching value between the query semantic identification information and the object semantic identification information according to the preset hierarchical matching information.

[0068] Specifically, the hierarchical matching information refers to the hierarchical matching strategy. This strategy is used to calculate the matching value between query semantic identifiers and object semantic identifiers at different levels. It measures the relevance between query information and candidate objects at different levels, so that the target object screening operation can be completed by combining the relevance at different levels. It should be noted that the hierarchical levels set in the hierarchical matching information can be set according to actual needs. For example, in an e-commerce scenario, three dimensions such as category, subcategory, and brand can be set to quickly complete the target object screening. Correspondingly, the identifier matching value refers to the comprehensive value between query semantic identifiers and object semantic identifiers obtained by combining the matching values ​​of each level. It represents the degree of matching of semantic IDs between the query and the object. The higher the value, the higher the similarity of the semantic IDs between the two. In specific implementation, the identifier matching value can be set in various forms, such as percentage values, high, medium, and low value levels, specific quantitative values, etc. This embodiment does not make any limitations here, and different evaluation standards can be set for different forms of values, so that the degree of matching between information can be reflected through specific values.

[0069] Based on this, after obtaining the query semantic identifier information corresponding to the object query information and the object semantic identifier information corresponding to each candidate object, in order to complete the high-precision object filtering operation by combining the semantic identifier information, the identifier information matching value between the query semantic identifier information and the object semantic identifier information can be calculated by combining the preset hierarchical matching information. This achieves the identification information matching value obtained by combining the values ​​between the query semantic identifier information and the object semantic identifier information in multiple levels, thereby facilitating the subsequent target object filtering operation and ensuring that the target object is more closely matched to the object query information.

[0070] Furthermore, when calculating the identifier matching value according to the hierarchical matching information, the calculation can be completed by combining the sub-values ​​corresponding to each information matching dimension. In this embodiment, determining the identifier matching value between the query semantic identifier information and the object semantic identifier information according to the preset hierarchical matching information includes: According to the preset hierarchical matching information, multiple information matching dimensions are determined; based on the query semantic identifier information and the object semantic identifier information, semantic identifier sub-information pairs associated with the multiple information matching dimensions are constructed, wherein the semantic identifier sub-information pairs contain query semantic identifier sub-information and object semantic identifier sub-information; based on the semantic identifier sub-information pairs, identifier information matching sub-values ​​corresponding to the multiple information matching dimensions are determined; the identifier information matching sub-values ​​corresponding to the multiple information matching dimensions are fused to obtain the identifier information matching value.

[0071] Specifically, multiple information matching dimensions refer to pre-defined information matching dimensions. It's important to note that these dimensions exhibit a gradient decreasing or increasing relationship, meaning the granularity of the calculated sub-values ​​gradually increases or decreases to ensure broader coverage and thus more accurate final identifier matching values. Correspondingly, semantic identifier sub-information pairs refer to information pairs constructed by combining query semantic identifier sub-information and object semantic identifier sub-information. Each pair consists of identifier sub-information from the query and object semantic identifier information, each containing the corresponding matching dimension. The identifier matching sub-value is the value of the corresponding matching dimension. The fusion processing can be understood as the summation, weighted calculation, or mean calculation of the identifier matching sub-values ​​corresponding to multiple information matching dimensions.

[0072] Based on this, when calculating the matching value of the identifier information, multiple information matching dimensions can be determined first according to the preset hierarchical matching information. Then, based on the query semantic identifier information and the object semantic identifier information, semantic identifier sub-information pairs associated with each of the multiple information matching dimensions can be constructed. The semantic identifier sub-information pairs contain query semantic identifier sub-information and object semantic identifier sub-information. After obtaining the semantic identifier sub-information pairs associated with each information matching dimension, the values ​​can be calculated sequentially from coarse-grained to fine-grained, that is, the identifier information matching sub-values ​​corresponding to each of the multiple information matching dimensions can be calculated based on the semantic identifier sub-information pairs. Finally, the identifier information matching sub-values ​​corresponding to each of the multiple information matching dimensions can be merged to obtain the identifier information matching value, which can then be used for target object filtering processing.

[0073] In other words, since semantic identifier information has a multi-level structure, such as the semantic ID corresponding to the object query information being {1, 101, 186}, with different values ​​representing different levels of characteristics, matching calculations can be performed sequentially from coarse-grained to fine-grained according to this characteristic. After calculating the matching degree between the query semantic ID and the product semantic ID at each level, the matching degrees of each level can be combined to determine the final matching value between the query information and the candidate object, so that downstream object filtering operations can be performed.

[0074] Following the previous example, after determining the semantic ID {1, 101, 186} corresponding to the product query information {Brand A computer case}, and the product semantic ID corresponding to each product in the candidate product set, the matching degree between each candidate product and the product query information at different granularities can be determined by calculating the matching degree level by level from coarse-grained to fine-grained. For example, the first level of the semantic ID represents the appliance dimension, the second level represents the subcategories under appliances, such as computers, televisions, hair dryers, etc., and the third level represents a brand. With this hierarchical structure, the matching degree between the product query information and each candidate product in the appliance dimension, the matching degree between the product query information and each candidate product in the computer dimension, and the matching degree between the product query information and each candidate product in the Brand A dimension can be calculated. After obtaining the matching degree corresponding to each of the three levels, the comprehensive matching degree value between the product query information and each product can be determined by weighted calculation, so as to carry out the target product filtering operation in the future.

[0075] In summary, by using a hierarchical approach to calculate the matching value, we can ensure that the calculated value more accurately represents the correlation between the query and the object, so as to facilitate subsequent filtering operations.

[0076] Step S108: Based on the identification information matching value, filter the target objects that match the object query information in the candidate object set.

[0077] Specifically, after determining the matching value between the query semantic identifier information and the object semantic identifier information, considering that the matching value can reflect the relevance between the object query information and each candidate object, target objects that match the object query information can be filtered from the candidate object set based on the matching value, so that downstream processing such as recommendations to users can be performed. Here, the target object is the object selected from the candidate object set that matches the object query information; its number can be set according to actual needs, and this embodiment does not impose any limitations.

[0078] Following the previous example, after obtaining the overall matching score between the product query information and each product, the top-k products can be filtered according to the overall matching score as the target products corresponding to the product query information {Brand A computer case}. For example, the final target products include the Brand A S1 model computer case, the Brand A S2 model computer case, ... the Brand A Sn model computer case. Then, the various models of Brand A computer cases can be displayed on the product browsing page for users to choose from.

[0079] Furthermore, to reduce computational complexity during target object screening, a method of progressively reducing the number of objects being calculated can be used. In this embodiment, the step of filtering target objects that match the object query information from the candidate object set based on the identifier information matching value includes: Based on the identification information matching value, initial candidate objects are filtered from the candidate object set to form an initial candidate object set. If the initial candidate object set does not meet the object feedback condition, the target identification information matching value between the query semantic identification information and the object semantic identification information corresponding to the initial candidate objects in the initial candidate object set is determined according to the hierarchical matching information. The target identification information matching value is used as the identification information matching value, and the initial candidate object set is used as the candidate object set. The step of filtering initial candidate objects from the candidate object set based on the identification information matching value and forming an initial candidate object set is performed. Until an initial candidate object set that meets the object feedback condition is determined, the initial candidate objects included in the initial candidate object set that meets the object feedback condition are used as target objects that match the object query information.

[0080] Specifically, the initial candidate object set refers to the set of candidate objects after preliminary screening based on the matching values ​​of the identifier information. The object feedback condition refers to the condition for detecting whether the initial candidate object set can be used as the feedback result of the object query information, such as the number of objects, the number of brands, the type, etc. in the set. This embodiment does not make any limitations here.

[0081] Based on this, to reduce computational complexity, target object filtering can be performed step-by-step. Specifically, initial candidate objects can be filtered from the candidate object set based on the identifier matching value, forming an initial candidate object set. At this point, it can be checked whether the initial candidate object set meets the object feedback conditions. If the initial candidate object set does not meet the object feedback conditions, further filtering is needed. Therefore, according to the hierarchical matching information, the target identifier matching value between the query semantic identifier information and the object semantic identifier information corresponding to the initial candidate objects in the initial candidate object set can be determined. The target identifier matching value can then be used as the identifier matching value, and the initial candidate object set can be used as the candidate object set. The step of filtering initial candidate objects from the candidate object set based on the identifier matching value and forming the initial candidate object set can be performed. This process continues until a certain computation cycle is completed. When an initial candidate object set that meets the object feedback conditions is determined, the initial candidate objects contained in the initial candidate object set that meets the object feedback conditions can be used as the target objects matching the object query information.

[0082] The information processing method provided in this embodiment, in order to provide more accurate relevance calculation through discretized semantic identifier information, first inputs the object query information into a semantic identifier generation model after obtaining the object query information. This enables the semantic identifier generation model to construct query semantic identifier information corresponding to the object query information, and simultaneously determines the object semantic identifier information corresponding to each candidate object in the candidate object set. After obtaining the semantic identifier information corresponding to the query information and the candidate objects respectively, the identifier information matching value between the query semantic identifier information and the object semantic identifier information can be determined according to the preset hierarchical matching information. The identifier information matching value can reflect the degree of association between the query information and each candidate object. Finally, based on the identifier information matching value, the target objects that match the object query information can be filtered from the candidate object set. This method improves the accuracy of relevance calculation between query information and candidate objects in object query scenarios by introducing semantic identifier information. Furthermore, the use of hierarchical matching information to determine the final value further ensures that the value accurately reflects the true relevance between the query information and the candidate objects. The selected target objects can accurately match the object query information, thus facilitating downstream service use.

[0083] See Figure 3 , Figure 3 A flowchart of another information processing method according to an embodiment of this specification is shown. The other information processing method is applied to a transaction server and specifically includes the following steps.

[0084] Step S302: Receive product query information submitted by the client through the product query page.

[0085] Step S304: Input the product query information into the semantic identifier generation model for processing to obtain the query semantic identifier information corresponding to the product query information, and determine the product semantic identifier information corresponding to each candidate product in the candidate product set.

[0086] Step S306: Determine the matching value of the identifier information between the query semantic identifier information and the product semantic identifier information according to the preset hierarchical matching information.

[0087] Step S308: Based on the identification information matching value, filter the target products that match the product query information in the candidate product set, and feed back the product information corresponding to the target products to the client.

[0088] This embodiment provides another information processing method. For any content not described in detail, please refer to the description in the above embodiments. This embodiment will not elaborate further here.

[0089] The information processing method provided in this embodiment, in order to provide more accurate relevance calculation through discrete semantic identifier information, first inputs the product query information into a semantic identifier generation model after obtaining the product query information. This enables the semantic identifier generation model to construct query semantic identifier information corresponding to the product query information, and simultaneously determines the product semantic identifier information corresponding to each product object in the candidate product set. After obtaining the semantic identifier information corresponding to the query information and the candidate products, the identifier information matching value between the query semantic identifier information and the product semantic identifier information can be determined according to the preset hierarchical matching information. The identifier information matching value can reflect the degree of association between the query information and each candidate product. Finally, based on the identifier information matching value, the target products that match the product query information can be filtered from the candidate product set, and the product information corresponding to the target products can be fed back to the client. This method improves the accuracy of relevance calculation between query information and candidate products in product query scenarios by introducing semantic identifier information. Furthermore, the use of hierarchical matching information to determine the final value further ensures that the value accurately reflects the true relevance between the query information and the candidate products. The selected target products can accurately match the product query information, thereby recommending the products that users need and meeting their shopping needs.

[0090] The following is in conjunction with the appendix Figure 4 Taking the application of the information processing method provided in this specification in a product query scenario as an example, the information processing method will be further explained. Among other things, Figure 4 A flowchart illustrating the processing steps of an information processing method provided in one embodiment of this specification is shown, specifically including the following steps.

[0091] Step S402: Receive product query information submitted by the client through the product query page.

[0092] Step S404: Input the product query information into the semantic identifier generation model, wherein the semantic identifier generation model includes an encoding module and a codebook quantization module.

[0093] Step S406: Use the encoding module to encode the product query information to obtain the query encoding vector corresponding to the product query information.

[0094] Step S408: Use the codebook quantization module to perform multi-layer residual quantization processing on the query encoding vector, and determine the query semantic identifier information corresponding to the product query information based on the processing result.

[0095] Step S410: Determine historical candidate products and newly added candidate products in the candidate product set, and extract the product semantic identifier information corresponding to the historical candidate products from the identifier information repository.

[0096] Step S412: Determine the product title information corresponding to the newly added candidate product, and input the product title information into the semantic identifier generation model for processing to obtain the product semantic identifier information corresponding to the newly added candidate product.

[0097] Step S414: Determine multiple information matching dimensions according to the preset hierarchical matching information.

[0098] Step S416: Based on the query semantic identifier information and the product semantic identifier information, construct multiple semantic identifier sub-information pairs that are associated with different information matching dimensions. The semantic identifier sub-information pairs contain query semantic identifier sub-information and product semantic identifier sub-information.

[0099] Step S418: Determine the identifier matching sub-values ​​corresponding to multiple information matching dimensions based on the semantic identifier sub-information.

[0100] Step S420: Merge the identification information matching sub-values ​​corresponding to multiple information matching dimensions to obtain the identification information matching value.

[0101] Step S422: Based on the identification information matching value, filter the target products that match the product query information from the candidate product set, and feed back the product information corresponding to the target products to the client.

[0102] The information processing method provided in this embodiment, in order to provide more accurate relevance calculation through discrete semantic identifier information, first inputs the product query information into a semantic identifier generation model after obtaining the product query information. This enables the semantic identifier generation model to construct query semantic identifier information corresponding to the product query information, and simultaneously determines the product semantic identifier information corresponding to each product object in the candidate product set. After obtaining the semantic identifier information corresponding to the query information and the candidate products, the identifier information matching value between the query semantic identifier information and the product semantic identifier information can be determined according to the preset hierarchical matching information. The identifier information matching value can reflect the degree of association between the query information and each candidate product. Finally, based on the identifier information matching value, the target products that match the product query information can be filtered from the candidate product set, and the product information corresponding to the target products can be fed back to the client. This method improves the accuracy of relevance calculation between query information and candidate products in product query scenarios by introducing semantic identifier information. Furthermore, the use of hierarchical matching information to determine the final value further ensures that the value accurately reflects the true relevance between the query information and the candidate products. The selected target products can accurately match the product query information, thereby recommending the products that users need and meeting their shopping needs.

[0103] Corresponding to the above method embodiments, this specification also provides information processing system embodiments. Figure 5 A schematic diagram of the structure of an information processing system according to one embodiment of this specification is shown. Figure 5 As shown, the information processing system 500 includes a client 510 and a server 520, comprising: The client 510 is used to determine object query information in response to a query request submitted through the object query page, and send the object query information to the server. The server 520 is used to input the object query information into the semantic identifier generation model for processing, obtain the query semantic identifier information corresponding to the object query information, and determine the object semantic identifier information corresponding to each candidate object in the candidate object set; determine the identifier information matching value between the query semantic identifier information and the object semantic identifier information according to the preset hierarchical matching information; based on the identifier information matching value, filter the target object matching the object query information in the candidate object set, and feed back the object information corresponding to the target object to the client.

[0104] In an optional embodiment, determining the identifier matching value between the query semantic identifier information and the object semantic identifier information according to preset hierarchical matching information includes: According to the preset hierarchical matching information, multiple information matching dimensions are determined; based on the query semantic identifier information and the object semantic identifier information, semantic identifier sub-information pairs associated with the multiple information matching dimensions are constructed, wherein the semantic identifier sub-information pairs contain query semantic identifier sub-information and object semantic identifier sub-information; based on the semantic identifier sub-information pairs, identifier information matching sub-values ​​corresponding to the multiple information matching dimensions are determined; the identifier information matching sub-values ​​corresponding to the multiple information matching dimensions are fused to obtain the identifier information matching value.

[0105] In an optional embodiment, the step of inputting the object query information into a semantic identifier generation model for processing to obtain query semantic identifier information corresponding to the object query information includes: The object query information is input into a semantic identifier generation model, wherein the semantic identifier generation model includes an encoding module and a codebook quantization module; the encoding module is used to encode the object query information to obtain a query encoding vector corresponding to the object query information; the codebook quantization module is used to perform multi-level residual quantization on the query encoding vector, and the query semantic identifier information corresponding to the object query information is determined based on the processing result.

[0106] In an optional embodiment, determining the object semantic identifier information corresponding to each candidate object in the candidate object set includes: Historical candidate objects and newly added candidate objects are determined from the candidate object set, and the semantic identification information of the objects corresponding to the historical candidate objects is extracted from the identification information repository; the object title information corresponding to the newly added candidate objects is determined, and the object title information is input into the semantic identification generation model for processing to obtain the object semantic identification information corresponding to the newly added candidate objects.

[0107] In an optional embodiment, training the semantic identifier generation model includes: Obtain query object sample pairs, wherein the query object sample pairs include sample query information and sample objects; use the query object sample pairs to perform a first-stage training on an initial semantic identifier generation model to obtain an intermediate semantic identifier generation model; use the query object sample pairs to perform a second-stage training on the intermediate semantic identifier generation model to obtain the semantic identifier generation model; wherein the first-stage training is used for the semantic identifier generation model to learn the construction of object semantic identifier information, and the second-stage training is used for the semantic identifier generation model to learn the construction of query semantic identifier information.

[0108] In an optional embodiment, the step of using the query object samples to perform a first-stage training on the initial semantic label generation model to obtain an intermediate semantic label generation model includes: The sample query information and the sample object are input into an initial semantic identifier generation model, which includes an initial encoding module and an initial codebook quantization module. The initial encoding module encodes the sample query information and the sample object respectively to obtain a sample query encoding vector corresponding to the sample query information and a sample object encoding vector corresponding to the sample object. The initial codebook quantization module performs multi-level residual quantization on the sample query encoding vector and the sample object encoding vector to obtain a discrete semantic identifier information sequence. Based on the first sample label corresponding to the query object sample and the discrete semantic identifier information sequence, the initial semantic identifier generation model is trained through comparative learning to obtain an intermediate semantic identifier generation model.

[0109] In an optional embodiment, the step of using the query object samples to perform a second-stage training on the intermediate semantic identifier generation model to obtain the semantic identifier generation model includes: Obtain the semantic identifier information of the sample objects associated with the first stage training, and use the semantic identifier information of the sample objects as the second sample label; input the sample query information into the intermediate semantic identifier generation model for processing to obtain the predicted object semantic identifier information sequence corresponding to the sample query information; perform generative training on the intermediate semantic identifier generation model based on the second sample label and the predicted object semantic identifier information sequence to obtain the semantic identifier generation model.

[0110] In an optional embodiment, the step of filtering target objects that match the object query information from the candidate object set based on the identification information matching value includes: Based on the identification information matching value, initial candidate objects are filtered from the candidate object set to form an initial candidate object set. If the initial candidate object set does not meet the object feedback condition, the target identification information matching value between the query semantic identification information and the object semantic identification information corresponding to the initial candidate objects in the initial candidate object set is determined according to the hierarchical matching information. The target identification information matching value is used as the identification information matching value, and the initial candidate object set is used as the candidate object set. The step of filtering initial candidate objects from the candidate object set based on the identification information matching value and forming an initial candidate object set is performed. Until an initial candidate object set that meets the object feedback condition is determined, the initial candidate objects included in the initial candidate object set that meets the object feedback condition are used as target objects that match the object query information.

[0111] The information processing method provided in this embodiment, in order to provide more accurate relevance calculation through discretized semantic identifier information, first inputs the object query information into a semantic identifier generation model after obtaining the object query information. This enables the semantic identifier generation model to construct query semantic identifier information corresponding to the object query information, and simultaneously determines the object semantic identifier information corresponding to each candidate object in the candidate object set. After obtaining the semantic identifier information corresponding to the query information and the candidate objects respectively, the identifier information matching value between the query semantic identifier information and the object semantic identifier information can be determined according to the preset hierarchical matching information. The identifier information matching value can reflect the degree of association between the query information and each candidate object. Finally, based on the identifier information matching value, the target objects that match the object query information can be filtered from the candidate object set. This method improves the accuracy of relevance calculation between query information and candidate objects in object query scenarios by introducing semantic identifier information. Furthermore, the use of hierarchical matching information to determine the final value further ensures that the value accurately reflects the true relevance between the query information and the candidate objects. The selected target objects can accurately match the object query information, thus facilitating downstream service use.

[0112] The above is an illustrative scheme of an information processing system according to this embodiment. It should be noted that the technical solution of this information processing system and the technical solution of the information processing method described above belong to the same concept. For details not described in detail in the technical solution of the information processing system, please refer to the description of the technical solution of the information processing method described above.

[0113] Corresponding to the above method embodiments, this specification also provides embodiments of an information processing apparatus. Figure 6 A schematic diagram of the structure of an information processing apparatus according to one embodiment of this specification is shown. Figure 6 As shown, the device includes: Module 602 is configured to retrieve object query information; The processing module 604 is configured to input the object query information into the semantic identifier generation model for processing, to obtain the query semantic identifier information corresponding to the object query information, and to determine the object semantic identifier information corresponding to each candidate object in the candidate object set. The determining module 606 is configured to determine the identifier matching value between the query semantic identifier information and the object semantic identifier information according to preset hierarchical matching information; The filtering module 608 is configured to filter target objects that match the object query information in the candidate object set based on the identification information matching value.

[0114] In an optional embodiment, determining the identifier matching value between the query semantic identifier information and the object semantic identifier information according to preset hierarchical matching information includes: According to the preset hierarchical matching information, multiple information matching dimensions are determined; based on the query semantic identifier information and the object semantic identifier information, semantic identifier sub-information pairs associated with the multiple information matching dimensions are constructed, wherein the semantic identifier sub-information pairs contain query semantic identifier sub-information and object semantic identifier sub-information; based on the semantic identifier sub-information pairs, identifier information matching sub-values ​​corresponding to the multiple information matching dimensions are determined; the identifier information matching sub-values ​​corresponding to the multiple information matching dimensions are fused to obtain the identifier information matching value.

[0115] In an optional embodiment, the step of inputting the object query information into a semantic identifier generation model for processing to obtain query semantic identifier information corresponding to the object query information includes: The object query information is input into a semantic identifier generation model, wherein the semantic identifier generation model includes an encoding module and a codebook quantization module; the encoding module is used to encode the object query information to obtain a query encoding vector corresponding to the object query information; the codebook quantization module is used to perform multi-level residual quantization on the query encoding vector, and the query semantic identifier information corresponding to the object query information is determined based on the processing result.

[0116] In an optional embodiment, determining the object semantic identifier information corresponding to each candidate object in the candidate object set includes: Historical candidate objects and newly added candidate objects are determined from the candidate object set, and the semantic identification information of the objects corresponding to the historical candidate objects is extracted from the identification information repository; the object title information corresponding to the newly added candidate objects is determined, and the object title information is input into the semantic identification generation model for processing to obtain the object semantic identification information corresponding to the newly added candidate objects.

[0117] In an optional embodiment, training the semantic identifier generation model includes: Obtain query object sample pairs, wherein the query object sample pairs include sample query information and sample objects; use the query object sample pairs to perform a first-stage training on an initial semantic identifier generation model to obtain an intermediate semantic identifier generation model; use the query object sample pairs to perform a second-stage training on the intermediate semantic identifier generation model to obtain the semantic identifier generation model; wherein the first-stage training is used for the semantic identifier generation model to learn the construction of object semantic identifier information, and the second-stage training is used for the semantic identifier generation model to learn the construction of query semantic identifier information.

[0118] In an optional embodiment, the step of using the query object samples to perform a first-stage training on the initial semantic label generation model to obtain an intermediate semantic label generation model includes: The sample query information and the sample object are input into an initial semantic identifier generation model, which includes an initial encoding module and an initial codebook quantization module. The initial encoding module encodes the sample query information and the sample object respectively to obtain a sample query encoding vector corresponding to the sample query information and a sample object encoding vector corresponding to the sample object. The initial codebook quantization module performs multi-level residual quantization on the sample query encoding vector and the sample object encoding vector to obtain a discrete semantic identifier information sequence. Based on the first sample label corresponding to the query object sample and the discrete semantic identifier information sequence, the initial semantic identifier generation model is trained through comparative learning to obtain an intermediate semantic identifier generation model.

[0119] In an optional embodiment, the step of using the query object samples to perform a second-stage training on the intermediate semantic identifier generation model to obtain the semantic identifier generation model includes: Obtain the semantic identifier information of the sample objects associated with the first stage training, and use the semantic identifier information of the sample objects as the second sample label; input the sample query information into the intermediate semantic identifier generation model for processing to obtain the predicted object semantic identifier information sequence corresponding to the sample query information; perform generative training on the intermediate semantic identifier generation model based on the second sample label and the predicted object semantic identifier information sequence to obtain the semantic identifier generation model.

[0120] In an optional embodiment, the step of filtering target objects that match the object query information from the candidate object set based on the identification information matching value includes: Based on the identification information matching value, initial candidate objects are filtered from the candidate object set to form an initial candidate object set. If the initial candidate object set does not meet the object feedback condition, the target identification information matching value between the query semantic identification information and the object semantic identification information corresponding to the initial candidate objects in the initial candidate object set is determined according to the hierarchical matching information. The target identification information matching value is used as the identification information matching value, and the initial candidate object set is used as the candidate object set. The step of filtering initial candidate objects from the candidate object set based on the identification information matching value and forming an initial candidate object set is performed. Until an initial candidate object set that meets the object feedback condition is determined, the initial candidate objects included in the initial candidate object set that meets the object feedback condition are used as target objects that match the object query information.

[0121] The information processing method provided in this embodiment, in order to provide more accurate relevance calculation through discretized semantic identifier information, first inputs the object query information into a semantic identifier generation model after obtaining the object query information. This enables the semantic identifier generation model to construct query semantic identifier information corresponding to the object query information, and simultaneously determines the object semantic identifier information corresponding to each candidate object in the candidate object set. After obtaining the semantic identifier information corresponding to the query information and the candidate objects respectively, the identifier information matching value between the query semantic identifier information and the object semantic identifier information can be determined according to the preset hierarchical matching information. The identifier information matching value can reflect the degree of association between the query information and each candidate object. Finally, based on the identifier information matching value, the target objects that match the object query information can be filtered from the candidate object set. This method improves the accuracy of relevance calculation between query information and candidate objects in object query scenarios by introducing semantic identifier information. Furthermore, the use of hierarchical matching information to determine the final value further ensures that the value accurately reflects the true relevance between the query information and the candidate objects. The selected target objects can accurately match the object query information, thus facilitating downstream service use.

[0122] The above is an illustrative scheme of an information processing device according to this embodiment. It should be noted that the technical solution of this information processing device and the technical solution of the information processing method described above belong to the same concept. For details not described in detail in the technical solution of the information processing device, please refer to the description of the technical solution of the information processing method described above.

[0123] Corresponding to the above method embodiments, this specification also provides another embodiment of an information processing apparatus. Figure 7 A schematic diagram of another information processing apparatus provided in one embodiment of this specification is shown. Figure 7As shown, this device is used on a transaction server and includes: The receiving module 702 is configured to receive product query information submitted by the client through the product query page; The processing module 704 is configured to input the product query information into the semantic identifier generation model for processing, obtain the query semantic identifier information corresponding to the product query information, and determine the product semantic identifier information corresponding to each candidate product in the candidate product set. The determination module 706 is configured to determine the identification information matching value between the query semantic identification information and the product semantic identification information according to preset hierarchical matching information; The feedback module 708 is configured to filter target products that match the product query information in the candidate product set based on the identification information matching value, and to feed back the product information corresponding to the target product to the client.

[0124] The information processing method provided in this embodiment, in order to provide more accurate relevance calculation through discrete semantic identifier information, first inputs the product query information into a semantic identifier generation model after obtaining the product query information. This enables the semantic identifier generation model to construct query semantic identifier information corresponding to the product query information, and simultaneously determines the product semantic identifier information corresponding to each product object in the candidate product set. After obtaining the semantic identifier information corresponding to the query information and the candidate products, the identifier information matching value between the query semantic identifier information and the product semantic identifier information can be determined according to the preset hierarchical matching information. The identifier information matching value can reflect the degree of association between the query information and each candidate product. Finally, based on the identifier information matching value, the target products that match the product query information can be filtered from the candidate product set, and the product information corresponding to the target products can be fed back to the client. This method improves the accuracy of relevance calculation between query information and candidate products in product query scenarios by introducing semantic identifier information. Furthermore, the use of hierarchical matching information to determine the final value further ensures that the value accurately reflects the true relevance between the query information and the candidate products. The selected target products can accurately match the product query information, thereby recommending the products that users need and meeting their shopping needs.

[0125] The above is an illustrative scheme of another information processing device according to this embodiment. It should be noted that the technical solution of this information processing device and the technical solution of the information processing method described above belong to the same concept. For details not described in detail in the technical solution of the information processing device, please refer to the description of the technical solution of the information processing method described above.

[0126] Figure 8A structural block diagram of a computing device 800 according to one embodiment of this specification is shown. The components of the computing device 800 include, but are not limited to, a memory 810 and a processor 820. The processor 820 is connected to the memory 810 via a bus 830, and a database 850 is used to store data.

[0127] The computing device 800 also includes an access device 840, which enables the computing device 800 to communicate via one or more networks 860. Examples of these networks include Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or combinations of communication networks such as the Internet. The access device 840 may include one or more of any type of wired or wireless network interface (e.g., a network interface card (NIC)), such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Wi-MAX (Worldwide Interoperability for Microwave Access) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, or a Near Field Communication (NFC) interface.

[0128] In one embodiment of this specification, the above-described components of the computing device 800 and Figure 8 Other components, not shown, can also be connected to each other, for example, via a bus. It should be understood that... Figure 8 The block diagram of the computing device shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0129] The computing device 800 can be any type of stationary or mobile computing device, including mobile computers or mobile computing devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smartphones), wearable computing devices (e.g., smartwatches, smart glasses, etc.) or other types of mobile devices, or stationary computing devices such as desktop computers or personal computers (PCs). The computing device 800 can also be a mobile or stationary server.

[0130] The processor 820 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above-described information processing method.

[0131] The above is an illustrative scheme of a computing device according to this embodiment. It should be noted that the technical solution of this computing device and the technical solution of the information processing method described above belong to the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the information processing method described above.

[0132] An embodiment of this specification also provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the above-described information processing method.

[0133] The above is an illustrative scheme of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium and the technical solution of the information processing method described above belong to the same concept. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the information processing method described above.

[0134] An embodiment of this specification also provides a computer program product, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described information processing method.

[0135] The above is an illustrative scheme of a computer program product according to this embodiment. It should be noted that the technical solution of this computer program product and the technical solution of the information processing method described above belong to the same concept. For details not described in detail in the technical solution of the computer program product, please refer to the description of the technical solution of the information processing method described above.

[0136] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0137] The computer program / instructions include computer program code, which may be in the form of source code, object code, executable file, or certain intermediate forms. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium may be appropriately added or removed according to the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0138] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments in this specification are not limited to the described order of actions, because according to the embodiments in this specification, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the embodiments in this specification.

[0139] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0140] The preferred embodiments disclosed above are merely illustrative of this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the embodiments described herein. These embodiments are selected and specifically described in this specification to better explain the principles and practical applications of the embodiments, thereby enabling those skilled in the art to better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. An information processing method, comprising: Retrieve object query information; The object query information is input into the semantic identifier generation model for processing to obtain the query semantic identifier information corresponding to the object query information, and to determine the object semantic identifier information corresponding to each candidate object in the candidate object set. Based on preset hierarchical matching information, determine the matching value of the identifier information between the query semantic identifier information and the object semantic identifier information; Based on the identification information matching value, target objects that match the object query information are filtered from the candidate object set.

2. The information processing method according to claim 1, wherein determining the identifier matching value between the query semantic identifier information and the object semantic identifier information according to preset hierarchical matching information includes: Based on the preset hierarchical matching information, determine multiple information matching dimensions; Based on the query semantic identifier information and the object semantic identifier information, construct semantic identifier sub-information pairs that are respectively associated with the multiple information matching dimensions, wherein the semantic identifier sub-information pairs include query semantic identifier sub-information and object semantic identifier sub-information; Based on the semantic identifier sub-information, determine the identifier information matching sub-values ​​corresponding to the multiple information matching dimensions respectively; The identification information matching sub-values ​​corresponding to the multiple information matching dimensions are merged to obtain the identification information matching value.

3. The information processing method according to claim 1, wherein inputting the object query information into a semantic identifier generation model for processing to obtain query semantic identifier information corresponding to the object query information includes: The object query information is input into the semantic identifier generation model, wherein the semantic identifier generation model includes an encoding module and a codebook quantization module; The object query information is encoded using the encoding module to obtain the query encoding vector corresponding to the object query information; The query encoding vector is subjected to multi-level residual quantization processing using the codebook quantization module, and the query semantic identifier information corresponding to the object query information is determined based on the processing result.

4. The information processing method according to claim 1, wherein determining the object semantic identifier information corresponding to each candidate object in the candidate object set includes: In the candidate object set, historical candidate objects and newly added candidate objects are determined, and the object semantic identification information corresponding to the historical candidate objects is extracted from the identification information repository. The object title information corresponding to the newly added candidate object is determined, and the object title information is input into the semantic identifier generation model for processing to obtain the object semantic identifier information corresponding to the newly added candidate object.

5. The information processing method according to claim 3, wherein training the semantic identifier generation model includes: Obtain a sample pair of query objects, wherein the sample pair of query objects includes sample query information and sample objects; The initial semantic identifier generation model is trained in the first stage using the query object samples to obtain an intermediate semantic identifier generation model. The intermediate semantic identifier generation model is trained in the second stage using the query object samples to obtain the semantic identifier generation model; The first stage of training is used for the semantic identifier generation model to learn the construction of object semantic identifier information, and the second stage of training is used for the semantic identifier generation model to learn the construction of query semantic identifier information.

6. The information processing method according to claim 5, wherein the step of using the query object sample pair to perform a first-stage training on the initial semantic identifier generation model to obtain an intermediate semantic identifier generation model includes: The sample query information and the sample object are input into the initial semantic identifier generation model, wherein the initial semantic identifier generation model includes an initial encoding module and an initial codebook quantization module; The initial encoding module is used to encode the sample query information and the sample object respectively to obtain the sample query encoding vector corresponding to the sample query information and the sample object encoding vector corresponding to the sample object. The initial codebook quantization module is used to perform multi-level residual quantization on the sample query encoding vector and the sample object encoding vector to obtain a discrete semantic identifier information sequence. Based on the first sample label corresponding to the query object sample pair and the discrete semantic identifier information sequence, the initial semantic identifier generation model is trained by comparative learning to obtain the intermediate semantic identifier generation model.

7. The information processing method according to claim 5 or 6, wherein the step of using the query object sample to perform a second-stage training on the intermediate semantic identifier generation model to obtain the semantic identifier generation model includes: Obtain the semantic identifier information of the sample objects associated with the first stage of training, and use the semantic identifier information of the sample objects as the second sample label; The sample query information is input into the intermediate semantic identifier generation model for processing to obtain the sequence of predicted object semantic identifier information corresponding to the sample query information; Based on the second sample label and the semantic identifier information sequence of the predicted object, the intermediate semantic identifier generation model is generatively trained to obtain the semantic identifier generation model.

8. The information processing method according to claim 1, wherein filtering target objects that match the object query information in the candidate object set based on the identification information matching value includes: Based on the matching value of the identification information, initial candidate objects are selected from the candidate object set and an initial candidate object set is formed. If the initial candidate object set does not meet the object feedback condition, the target identifier information matching value between the query semantic identifier information and the object semantic identifier information corresponding to the initial candidate object in the initial candidate object set is determined according to the hierarchical matching information. The target identification information matching value is used as the identification information matching value, the initial candidate object set is used as the candidate object set, and the steps of filtering initial candidate objects in the candidate object set based on the identification information matching value and forming an initial candidate object set are performed. Once an initial set of candidate objects that meets the object feedback conditions is determined, the initial candidate objects contained in the initial set of candidate objects that meet the object feedback conditions are taken as target objects that match the object query information.

9. An information processing method, applied to a transaction server, comprising: Receive product query information submitted by the client through the product query page; The product query information is input into the semantic identifier generation model for processing to obtain the query semantic identifier information corresponding to the product query information, and to determine the product semantic identifier information corresponding to each candidate product in the candidate product set. Based on preset hierarchical matching information, determine the matching value of the identifier information between the query semantic identifier information and the product semantic identifier information; Based on the identification information matching value, target products that match the product query information are selected from the candidate product set, and the product information corresponding to the target products is fed back to the client.

10. An information processing system, comprising a client and a server, including: The client is configured to determine object query information in response to a query request submitted through the object query page, and send the object query information to the server. The server is configured to input the object query information into a semantic identifier generation model for processing, obtain the query semantic identifier information corresponding to the object query information, and determine the object semantic identifier information corresponding to each candidate object in the candidate object set; determine the identifier information matching value between the query semantic identifier information and the object semantic identifier information according to preset hierarchical matching information; based on the identifier information matching value, filter the target objects that match the object query information in the candidate object set, and feed back the object information corresponding to the target objects to the client.

11. A computing device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 9.

12. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 9.

13. A computer program product comprising a computer program or instructions which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 9.

Images