Supplier data distribution method and device, electronic equipment and storage medium

Abstract

The invention provides a supplier data distribution method and device, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence, intelligent supply chains and big data. The method comprises the steps of obtaining business data and multiple supplier data; respectively coding structured fields and text description fields in the business data through a first coding model, and carrying out fusion processing to obtain business feature vectors; respectively coding and fusing structured fields and text description fields in the supplier data through a second coding model to obtain supplier feature vectors corresponding to the suppliers; determining similarity according to the business feature vector and the plurality of supplier feature vectors; and determining a target supplier according to the similarity. Different coding algorithms are set for structured fields and text description fields, deep semantic analysis of unstructured data is achieved, the feature extraction capacity of multi-source heterogeneous information is improved, multi-dimensional consideration is conducted on service data distribution, and the information utilization rate, reliability and service adaptability are improved.

Description

Technical Field

[0001] This disclosure relates to the fields of artificial intelligence, smart supply chain and big data technology, and in particular to a supplier data allocation method, apparatus, electronic device and storage medium. Background Technology

[0002] With the development of computer technology, business operations are becoming increasingly complex, and the amount of information that needs to be considered when allocating business tasks is also increasing.

[0003] In related technologies, once a system receives a business opportunity, it needs to allocate the corresponding business to at least one supplier. In this process, the system often relies on manually constructed single-dimensional business features, such as industry classification and procurement scale, making it difficult to effectively integrate heterogeneous data from different channels and formats. This results in low information utilization and a lack of multi-dimensional feature consideration. Furthermore, at the algorithm level, traditional collaborative filtering and decision tree models struggle to capture the non-linear relationships between high-dimensional features, especially when dealing with cross-domain heterogeneous data from both the buyer and supplier sides. Their modeling capabilities are significantly insufficient, thus reducing allocation accuracy and business adaptability.

[0004] Therefore, there is an urgent need for a supplier data allocation method to fully consider business data allocation from multiple dimensions, thereby improving the reliability and business adaptability of data allocation.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] This disclosure provides a supplier data allocation method, apparatus, electronic device, and storage medium, which at least to a certain extent fully considers the multi-dimensional aspects of business data allocation, thereby improving the reliability and business adaptability of data allocation.

[0007] Other features and advantages of this disclosure will become apparent from the following detailed description, or may be learned in part from practice of this disclosure.

[0008] According to one aspect of this disclosure, a supplier data allocation method is provided, comprising: acquiring business data and supplier data corresponding to multiple suppliers to be allocated; encoding structured fields and text description fields in the business data respectively using a first encoding model, and fusing the encoded first structured features and first text features to obtain a business feature vector; encoding structured fields and text description fields in the supplier data respectively using a second encoding model, and fusing the encoded second structured features and second text features to obtain a supplier feature vector corresponding to each supplier; determining multiple similarities based on the business feature vectors and multiple supplier feature vectors, wherein each business feature vector corresponds to a similarity with any supplier feature vector; and determining a target supplier from the multiple suppliers to be allocated based on the similarities.

[0009] In some exemplary embodiments, determining the target supplier from a plurality of suppliers to be assigned based on the similarity includes: determining N candidate suppliers from a plurality of suppliers to be assigned based on the similarity, where N is a positive integer; generating a rating result for each of the N candidate suppliers based on at least one of the similarity, supplier performance data, and urgency matching data, wherein the supplier performance data is used to describe the supplier's historical performance information, and the urgency matching data is used to describe the matching degree between the supplier's historical response time and the current business urgency; and determining at least one candidate supplier as the target supplier based on the rating result.

[0010] In some exemplary embodiments, the step of encoding the structured fields and text description fields in the business data using a first encoding model includes: encoding the structured fields in the business data using a fully connected layer in the first encoding model to obtain a first structured feature; and encoding the text description fields in the business data using a code encoder in the first encoding model to obtain a first text feature. The step of fusing the encoded first structured feature and the first text feature to obtain a business feature vector includes: fusing the first structured feature and the first text feature based on a multi-head attention fusion algorithm to obtain a business feature vector.

[0011] In some exemplary embodiments, the step of encoding the structured fields and text description fields in the supplier data using the second encoding model includes: encoding the text description fields in the supplier data using a text encoder in the second encoding model to obtain a second structured feature; and encoding the structured fields in the supplier data using a statistical encoder in the second encoding model to obtain a second text feature. The step of fusing the encoded second structured features and second text features to obtain the supplier feature vectors corresponding to each supplier includes: fusing the second structured features and second text features based on a dynamic gating network to obtain the supplier feature vectors.

[0012] In some exemplary embodiments, before obtaining the business data and the supplier data corresponding to the multiple suppliers to be assigned, the method further includes: obtaining a set of positive and negative samples, wherein positive samples are combinations of historically successfully matched businesses and suppliers, and negative samples are combinations of unmatched and / or uncompleted businesses and suppliers; and training the first encoding model and the second encoding model using the set of positive and negative samples.

[0013] In some exemplary embodiments, training the first encoding model and the second encoding model using the positive and negative sample set includes: performing two-stage training on the first encoding model and the second encoding model, wherein the first stage training is training after freezing the target parameters in the first encoding model and the second encoding model, and the second stage training is training the first encoding model and the second encoding model with all parameters.

[0014] In some exemplary embodiments, training the first encoding model and the second encoding model using the set of positive and negative samples includes: training the first encoding model and the second encoding model according to the information noise contrast estimation loss function, wherein the information noise contrast estimation loss function is obtained by adjusting the sample similarity distribution.

[0015] According to another aspect of this disclosure, a supplier data allocation apparatus is also provided, comprising: a data acquisition module for acquiring business data and supplier data corresponding to multiple suppliers to be allocated; a business feature vector determination module for encoding structured fields and text description fields in the business data respectively through a first encoding model, and fusing the encoded first structured feature and first text feature to obtain a business feature vector; a supplier feature vector determination module for encoding structured fields and text description fields in the supplier data respectively through a second encoding model, and fusing the encoded second structured feature and second text feature to obtain a supplier feature vector corresponding to each supplier; a similarity determination module for determining multiple similarities based on the business feature vector and multiple supplier feature vectors, wherein the business feature vector corresponds to a similarity with any supplier feature vector; and a target supplier determination module for determining a target supplier from the multiple suppliers to be allocated based on the similarity.

[0016] In some exemplary embodiments, the target supplier determination module is configured to determine N candidate suppliers from a plurality of suppliers to be assigned, where N is a positive integer, based on the similarity; generate a score result for each of the N candidate suppliers based on at least one of the similarity, supplier performance data, and urgency matching data, wherein the supplier performance data is used to describe the supplier's historical performance information, and the urgency matching data is used to describe the matching degree between the supplier's historical response time and the current business urgency; and determine at least one candidate supplier as the target supplier based on the score result.

[0017] In some exemplary embodiments, the business feature vector determination module is used to encode the structured fields in the business data through the fully connected layer in the first encoding model to obtain a first structured feature; to encode the text description fields in the business data through the code encoder in the first encoding model to obtain a first text feature; and to fuse the first structured feature and the first text feature based on a multi-head attention fusion algorithm to obtain a business feature vector.

[0018] In some exemplary embodiments, the supplier feature vector determination module is used to encode the text description field in the supplier data to obtain a second structured feature by using the text encoder in the second encoding model; to encode the structured field in the supplier data to obtain a second text feature by using the statistical encoder in the second encoding model; and to fuse the second structured feature and the second text feature based on a dynamic gating network to obtain a supplier feature vector.

[0019] In some exemplary embodiments, the apparatus provided in this disclosure further includes: The model training module is used to obtain a set of positive and negative samples, where positive samples are combinations of historically successfully matched businesses and suppliers, and negative samples are combinations of unmatched and / or unsuccessful businesses and suppliers; the first encoding model and the second encoding model are trained using the set of positive and negative samples.

[0020] In some exemplary embodiments, the model training module is used to perform two-stage training on the first encoding model and the second encoding model, wherein the first stage training is to train after freezing the target parameters in the first encoding model and the second encoding model, and the second stage training is to train the first encoding model and the second encoding model with all parameters.

[0021] In some exemplary embodiments, the model training module is used to train the first encoding model and the second encoding model according to the information noise contrast estimation loss function, wherein the information noise contrast estimation loss function is obtained by adjusting the sample similarity distribution.

[0022] According to another aspect of this disclosure, an electronic device is also provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform a supplier data allocation method of any of the above via executing the executable instructions.

[0023] According to another aspect of this disclosure, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the supplier data allocation method of any of the above.

[0024] According to another aspect of this disclosure, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the supplier data allocation method provided in various alternative embodiments of this disclosure.

[0025] The technical solutions provided in this disclosure can achieve deep semantic parsing of unstructured data by setting different encoding algorithms for structured fields and text description fields, thereby improving the feature extraction capability of multi-source heterogeneous information. As a result, this disclosure can fully consider the allocation of business data from multiple dimensions, improve information utilization, and enhance the reliability and business adaptability of data allocation.

[0026] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0027] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0028] Figure 1 A schematic diagram of a system architecture according to an embodiment of this disclosure is shown; Figure 2 This diagram illustrates a flowchart of a supplier data allocation method according to an embodiment of the present disclosure; Figure 3 This diagram illustrates the structure of a dual-tower matching model according to an embodiment of the present disclosure. Figure 4 This diagram illustrates a model training process according to an embodiment of the present disclosure. Figure 5 This diagram illustrates a process for determining a target supplier according to an embodiment of the present disclosure. Figure 6 This diagram illustrates a supplier data allocation device according to an embodiment of the present disclosure. Figure 7 This diagram illustrates a structural block diagram of an electronic device according to an embodiment of the present disclosure. Figure 8 A schematic diagram of a computer-readable storage medium according to an embodiment of the present disclosure is shown. Detailed Implementation

[0029] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that this disclosure will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0030] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0031] The specific implementation methods of the embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0032] Figure 1 A schematic diagram of an exemplary application system architecture to which the supplier data allocation method of the embodiments of this disclosure can be applied is shown. Figure 1 As shown, the system architecture may include terminal device 101, network 102 and server 103.

[0033] For example, terminal device 101 can send business data and supplier data corresponding to multiple suppliers to be assigned to server 103.

[0034] Then, the server 103 can obtain business data and supplier data corresponding to multiple suppliers to be assigned; encode the structured fields and text description fields in the business data respectively through a first encoding model, and fuse the first structured feature and the first text feature obtained by encoding to obtain a business feature vector; encode the structured fields and text description fields in the supplier data respectively through a second encoding model, and fuse the second structured feature and the second text feature obtained by encoding to obtain a supplier feature vector corresponding to each supplier; determine multiple similarities based on the business feature vector and multiple supplier feature vectors, wherein the business feature vector corresponds to a similarity with any supplier feature vector; and determine the target supplier from multiple suppliers to be assigned based on the similarity.

[0035] Finally, server 103 can return the identified target supplier to terminal device 101.

[0036] Network 102 is a medium used to provide a communication link between terminal device 101 and server 103, and can be a wired network or a wireless network.

[0037] Optionally, the aforementioned wireless or wired networks use standard communication technologies and / or protocols. The network is typically the Internet, but can also be any network, including but not limited to Local Area Networks (LANs), Metropolitan Area Networks (MANs), Wide Area Networks (WANs), mobile, wired or wireless networks, private networks, or any combination of virtual private networks. In some embodiments, technologies and / or formats, including Hyper Text Markup Language (HTML), Extensible Markup Language (XML), etc., are used to represent data exchanged over the network. Furthermore, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Networks (VPNs), and Internet Protocol Security (IPSec) can be used to encrypt all or some links. In other embodiments, custom and / or dedicated data communication technologies can be used to replace or supplement the aforementioned data communication technologies.

[0038] Terminal device 101 can be various electronic devices, including but not limited to smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, wearable devices, augmented reality devices, virtual reality devices, etc.

[0039] Server 103 can be a server that provides various services, such as a backend management server that supports the device operated by the user using terminal device 101. The backend management server can analyze and process received requests and other data, and feed the processing results back to the terminal device.

[0040] Optionally, the server can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0041] Those skilled in the art will know that Figure 1 The number of terminal devices 101, networks 102, and servers 103 shown is merely illustrative. Any number of terminal devices 101, networks 102, and servers 103 can be used as needed. This disclosure does not limit the number of such devices.

[0042] Under the above system architecture, this disclosure provides a supplier data allocation method, which can be executed by any electronic device with computing power.

[0043] In some embodiments, the supplier data allocation method provided in this disclosure can be executed by a terminal device of the system architecture described above; in other embodiments, the data allocation method provided in this disclosure can be executed by a server in the system architecture described above; in still other embodiments, the supplier data allocation method provided in this disclosure can be implemented by the terminal device and the server in the system architecture described above through interaction.

[0044] Figure 2 This invention discloses a flowchart of a supplier data allocation method according to an embodiment of the present disclosure, as shown below. Figure 2 As shown, the supplier data allocation method provided in this embodiment includes the following steps S202 to S210.

[0045] S202, Obtain business data and supplier data corresponding to multiple suppliers to be assigned.

[0046] This disclosure does not limit the content of the business data. For example, the business data may include a text description of a specific business order, such as a garment manufacturing order or a chemical product order. Additionally, the business data may also include the industry code, budget range, and urgency level corresponding to the business. It should be noted that the industry code represents the industry corresponding to the business order, the budget range represents the budget for the business order, and the urgency level indicates the degree of urgency at which the business order needs to be delivered or distributed.

[0047] Furthermore, this disclosure does not limit the content of the supplier data corresponding to the supplier. For example, for any supplier, the corresponding supplier data may include information on the services that can be provided, qualification documents, and historical transaction statistics, such as the success rate and response speed.

[0048] Therefore, embodiments of this disclosure can determine which suppliers to allocate the business order to based on business data and supplier data corresponding to multiple suppliers to be allocated.

[0049] S204, the structured fields and text description fields in the business data are encoded separately using the first encoding model, and the first structured feature and the first text feature obtained by encoding are fused to obtain the business feature vector.

[0050] In some embodiments, structured fields in the business data may include, for example, the industry code corresponding to the business, the budget range, and the urgency level. Text description fields in the business data may include, for example, text descriptions used to describe a particular business order.

[0051] In some exemplary embodiments, the structured fields and text description fields in the business data are encoded by a first encoding model, including: encoding the structured fields in the business data to obtain a first structured feature by using a fully connected layer in the first encoding model; and encoding the text description fields in the business data to obtain a first text feature by using a code encoder in the first encoding model.

[0052] This disclosure does not limit the code encoder. For example, the code encoder may be a CodeBERT (Code Bidirectional Encoder Representations from Transformers) encoder. Alternatively, the code encoder may be a BERT encoder, etc.

[0053] For example, the first encoding model may include one or more fully connected layers, since the fully connected layers can be used to perform linear transformations to obtain the corresponding first structured features.

[0054] In an exemplary embodiment, the business data can be preprocessed before being input into the first encoding model for encoding.

[0055] In some exemplary embodiments, the first structured feature and the first text feature obtained by encoding are fused to obtain a business feature vector, including: fusing the first structured feature and the first text feature based on a multi-head attention fusion algorithm to obtain a business feature vector.

[0056] For example, embodiments of this disclosure can use a multi-head attention mechanism to perform cross-modal fusion of the first text features encoded by the CodeBERT encoder and the first structured features obtained by the fully connected layer to obtain a business feature vector.

[0057] In one possible embodiment, the business feature vector can be a 768-dimensional vector.

[0058] S206, the structured fields and text description fields in the supplier data are encoded separately using the second encoding model, and the second structured features and second text features obtained by encoding are fused to obtain the supplier feature vector corresponding to each supplier.

[0059] In an exemplary embodiment, the structured fields in the supplier data may include historical transaction statistics, such as transaction rate and response speed. The text description fields in the supplier data may include at least one of service introduction information and qualification documents.

[0060] In some exemplary embodiments, the structured fields and text description fields in the supplier data are encoded by a second encoding model, including: encoding the text description fields in the supplier data to obtain a second structured feature by using a text encoder in the second encoding model; and encoding the structured fields in the supplier data to obtain a second text feature by using a statistical encoder in the second encoding model.

[0061] For example, when the text description field in the supplier data includes service introduction information, this service introduction information can be encoded using a text encoder. This text encoder could be, for example, a CodeBERT encoder, a BERT encoder, etc. Conversely, when the text description field in the supplier data includes qualification documents, these qualification documents can be parsed using a file parser.

[0062] It should be noted that the qualification document may be in PDF (Portable Document Format) or similar format, and this disclosure does not limit it to this.

[0063] For example, when the text description field in the supplier data includes service introduction information and qualification documents, the second text feature can be obtained through the output of the text encoder and the document parser.

[0064] In some exemplary embodiments, the encoded second structured features and second text features are fused to obtain supplier feature vectors corresponding to each supplier, including: fusing the second structured features and second text features based on a dynamic gating network to obtain supplier feature vectors.

[0065] For example, embodiments of this disclosure can use a dynamic gating network to dynamically weight and fuse a second text feature obtained by encoding at least one of a text encoder and a file parser, and a second structured feature obtained by encoding a statistical encoder, to obtain a supplier feature vector.

[0066] Alternatively, in other embodiments, the present disclosure embodiments may also achieve dynamic weighted fusion of the second text features and the second structured features through a multi-head attention fusion algorithm, and the present disclosure embodiments do not limit this.

[0067] In one possible embodiment, the supplier feature vector can be a 768-dimensional vector. That is, the dimension of the business feature vector can be the same as the dimension of the supplier feature vector.

[0068] S208. Based on the business feature vector and multiple supplier feature vectors, determine multiple similarities, where each business feature vector corresponds to a similarity with any supplier feature vector.

[0069] In an exemplary embodiment, the cosine similarity between the business feature vector and any supplier feature vector can be calculated to determine the degree of fit between the supplier and the business data. Therefore, embodiments of this disclosure can calculate the similarity between the business feature vector and each supplier feature vector separately.

[0070] S210, Based on similarity, determine the target supplier from multiple suppliers to be assigned.

[0071] In an exemplary embodiment, suppliers can be sorted from highest to lowest similarity based on the similarity between the business data and the data of each supplier. Then, it can be determined how many suppliers the current business needs to match. Taking A suppliers as an example, the A suppliers with the highest similarity can be selected as target suppliers, where A is a positive integer.

[0072] In other embodiments, the suppliers to be assigned can be dynamically reordered based on similarity to determine the target supplier.

[0073] In some exemplary embodiments, determining a target supplier from a plurality of suppliers to be assigned based on similarity includes: determining N candidate suppliers from a plurality of suppliers to be assigned based on similarity, where N is a positive integer; generating a score result for each of the N candidate suppliers based on at least one of similarity, supplier performance data, and urgency matching data, wherein the supplier performance data is used to describe the supplier's historical performance information, and the urgency matching data is used to describe the matching degree between the supplier's historical response time and the current business urgency; and determining at least one candidate supplier as the target supplier based on the score results.

[0074] In an exemplary embodiment, suppliers can be sorted from highest to lowest similarity based on the similarity between the business data and the data of each supplier. Then, the N suppliers with the highest similarity can be selected as candidate suppliers.

[0075] In some embodiments, the scoring results of N alternative suppliers can be generated according to the following formula (1) to achieve dynamic re-sorting.

[0076]

[0077] In formula (1), F can represent the scoring result. This can represent the corresponding similarity. 'x' can represent supplier fulfillment data. 'y' can represent urgency matching data. , , These can be used to represent the corresponding weighting coefficients.

[0078] It should be noted that... , , The value can be set based on experience or application scenarios, and this disclosed embodiment does not limit it in this way. Furthermore, the... , , The value of can also be adjusted periodically through reinforcement learning.

[0079] In an exemplary embodiment, the supplier performance data can be obtained based on statistics of the supplier's historical performance; for example, the better the historical performance, the higher the value of the supplier performance data. The urgency matching data can be determined based on the supplier's historical average response time and the urgency level corresponding to the business. For example, the better the supplier's historical average response time matches the urgency level corresponding to the business, the higher the value of the urgency matching data.

[0080] In some exemplary embodiments, the urgency matching data can be calculated according to the following formula (2).

[0081]

[0082] In formula (1), y can represent urgency matching data. T can represent the supplier's historical average response time. It can indicate the urgency of a business transaction.

[0083] In an exemplary embodiment, after determining the scoring results, the candidate suppliers can be ranked according to the scoring results, and at least one candidate supplier with the highest score can be selected as the target supplier. This disclosure does not limit the number of target suppliers.

[0084] For example, after ranking the candidate suppliers according to the scoring results, diversity control is performed. For instance, diversity control conditions may include requiring suppliers matching the same business to cover at least X regions, where X is a positive integer. Thus, this disclosure can determine target suppliers based on the scoring results and diversity control conditions.

[0085] It should be noted that the embodiments of this disclosure realize the dynamic utilization of the supplier's historical behavior, thereby effectively optimizing the matching results based on the supplier's actual performance, affecting the business adaptability of the final distribution and the quality of the supplier's response.

[0086] In some exemplary embodiments, before acquiring business data and supplier data corresponding to multiple suppliers to be assigned, the method provided in this disclosure may further include: acquiring a set of positive and negative samples, wherein positive samples are combinations of historically successfully matched businesses and suppliers, and negative samples are combinations of unmatched and / or unsuccessful transactions of businesses and suppliers; and training a first encoding model and a second encoding model using the set of positive and negative samples.

[0087] In an exemplary embodiment, the set of positive and negative samples may include multiple positive sets and multiple negative sets. The positive sets may be historically successfully matched <business, supplier> pairs. The negative samples may be obtained based on a dynamic negative sampling strategy. For example, the negative samples may include at least one of random negative samples and difficult negative samples. The random negative samples may be <business, supplier> pairs from the same industry that were not successfully matched.

[0088] It should be noted that if the similarity between the business feature vector corresponding to a certain business and the supplier feature vector corresponding to a certain supplier meets the first matching condition, then the business and the supplier are successfully matched; otherwise, the business and the supplier are not successfully matched.

[0089] Additionally, difficult negative samples can be <business, supplier> pairs with a similarity of not less than a first threshold, but which have not resulted in a transaction. This first threshold can be, for example, 0.6, 0.7, 0.8, etc., and this disclosure does not limit it.

[0090] In some exemplary embodiments, training the first encoding model and the second encoding model using a set of positive and negative samples includes: performing two-stage training on the first encoding model and the second encoding model, wherein the first stage of training is to train after freezing the target parameters in the first encoding model and the second encoding model, and the second stage of training is to train the first encoding model and the second encoding model with all parameters.

[0091] In an exemplary implementation, the CodeBERT parameters in the model can be trained first in the first training stage to train the feature fusion layer. The second training stage can involve fine-tuning the full parameters of both the first and second encoding models.

[0092] In some exemplary embodiments, training the first encoding model and the second encoding model using a set of positive and negative samples includes: training the first encoding model and the second encoding model according to the information noise contrast estimation loss function, wherein the information noise contrast estimation loss function is obtained by adjusting the sample similarity distribution.

[0093] In some embodiments, the information noise contrast estimation loss function can be as shown in the following formula (3).

[0094]

[0095] In formula (3), It can represent the business feature vector corresponding to business data. Additionally, It can represent the supplier feature vector corresponding to the suppliers that can form a positive set with this business. It can represent the supplier feature vector corresponding to the suppliers that can form a negative set with this business.

[0096] In addition, L in formula (3) can represent the information noise contrast estimation loss function. It can represent a dynamic temperature coefficient, which can be automatically adjusted based on the similarity distribution of batch samples.

[0097] The method provided in this disclosure, by setting different encoding algorithms for structured fields and text description fields, can achieve deep semantic parsing of unstructured data and improve the feature extraction capability of multi-source heterogeneous information. Thus, this disclosure can fully consider business data allocation from multiple dimensions, improve information utilization, and enhance the reliability and business adaptability of data allocation.

[0098] It should be noted that the acquisition, storage, use, and processing of data in this disclosure comply with relevant regulations. The various types of data obtained in this disclosure, such as personal identity data, operational data, and behavioral data related to individuals, customers, and groups, have all been licensed.

[0099] Figure 3 A schematic diagram of a dual-tower matching model provided in an embodiment of this disclosure is shown.

[0100] like Figure 3 As shown, the dual-tower matching model can include a first encoding model and a second encoding model. The text description field and structured field from the business data can be input into the CodeBERT encoder and fully connected layer of the first encoding model, respectively, to obtain the first text feature and the first structured feature. Then, the first structured feature and the first text feature can be fused using a multi-head attention fusion algorithm to obtain a 768-dimensional business feature vector.

[0101] In addition, in Figure 3 Furthermore, structured fields from supplier data, such as service descriptions, qualification documents, and historical transaction statistics (e.g., conversion rates, response times), can be input into the text encoder, file parser, and statistical encoder of the first encoding model. Then, feature fusion processing can be performed based on a dynamic gating network to obtain a 768-dimensional supplier feature vector.

[0102] Finally, cosine similarity can be calculated between the business feature vector and the supplier feature vector.

[0103] Figure 4 A schematic diagram of a model training process provided in an embodiment of this disclosure is shown.

[0104] like Figure 4 As shown, this embodiment of the disclosure can dynamically negative sample the training data to obtain positive and negative samples. Then, similarity calculation and loss function calculation can be performed based on the positive and negative samples to adjust the model parameters. Finally, backpropagation of parameters and updating of model weights can be performed. The above process is executed iteratively, increasing the difficulty of the training samples in stages according to the business complexity, thereby completing the training of the model.

[0105] Figure 5 A schematic diagram illustrating a process for determining a target supplier provided by an embodiment of this disclosure is shown.

[0106] like Figure 5As shown, when a new business has a target supplier to be assigned, the suppliers can be initially screened based on cosine similarity to determine candidate suppliers. For example, the top 50 suppliers with the highest similarity can be selected as candidate suppliers. Then, the candidate suppliers can be dynamically reordered, and the final target supplier can be determined based on diversity control. The final distribution list is then output.

[0107] For example, in the Figure 5 In this process, a weighted formula can be used to calculate a score based on cosine similarity, supplier performance data, and urgency matching data. This score can then be used to re-rank the suppliers to determine the final target suppliers.

[0108] Based on the same inventive concept, this disclosure also provides a supplier data allocation device, as described in the following embodiments. Since the principle by which this device solves the problem is similar to that of the method embodiments described above, the implementation of this device embodiment can refer to the implementation of the method embodiments described above, and repeated details will not be repeated.

[0109] Figure 6 This diagram illustrates a supplier data allocation device according to an embodiment of the present disclosure, such as... Figure 6 As shown, the device includes: The data acquisition module 601 is used to acquire business data and supplier data corresponding to multiple suppliers to be assigned; The business feature vector determination module 602 is used to encode the structured fields and text description fields in the business data respectively through the first encoding model, and to fuse the first structured feature and the first text feature obtained by encoding to obtain the business feature vector. The supplier feature vector determination module 603 is used to encode the structured fields and text description fields in the supplier data through the second encoding model, and to fuse the encoded second structured features and second text features to obtain the supplier feature vectors corresponding to each supplier. The similarity determination module 604 is used to determine multiple similarities based on the business feature vector and multiple supplier feature vectors, wherein the business feature vector corresponds to one similarity with any supplier feature vector; The target supplier determination module 605 is used to determine the target supplier from a plurality of suppliers to be assigned based on similarity.

[0110] In some exemplary embodiments, the target supplier determination module 605 is configured to determine N candidate suppliers from a plurality of suppliers to be assigned based on similarity, where N is a positive integer; generate a score result for each of the N candidate suppliers based on at least one of similarity, supplier performance data, and urgency matching data, wherein the supplier performance data is used to describe the supplier's historical performance information, and the urgency matching data is used to describe the matching degree between the supplier's historical response time and the current business urgency; and determine at least one candidate supplier as the target supplier based on the score results.

[0111] In some exemplary embodiments, the business feature vector determination module 602 is used to encode the structured fields in the business data through the fully connected layer in the first encoding model to obtain the first structured feature; to encode the text description fields in the business data through the code encoder in the first encoding model to obtain the first text feature; and to fuse the first structured feature and the first text feature based on a multi-head attention fusion algorithm to obtain the business feature vector.

[0112] In some exemplary embodiments, the supplier feature vector determination module 603 is used to encode the text description field in the supplier data to obtain a second structured feature by using a text encoder in the second encoding model; to encode the structured field in the supplier data to obtain a second text feature by using a statistical encoder in the second encoding model; and to fuse the second structured feature and the second text feature based on a dynamic gating network to obtain a supplier feature vector.

[0113] In some exemplary embodiments, the apparatus provided in this disclosure further includes: The model training module is used to obtain a set of positive and negative samples, where positive samples are combinations of successfully matched businesses and suppliers in the past, and negative samples are combinations of unmatched and / or unsuccessful businesses and suppliers; the first encoding model and the second encoding model are trained using the set of positive and negative samples.

[0114] In some exemplary embodiments, the model training module is used to perform two-stage training on the first encoding model and the second encoding model, wherein the first stage training is to train after freezing the target parameters in the first encoding model and the second encoding model, and the second stage training is to train the first encoding model and the second encoding model with all parameters.

[0115] In some exemplary embodiments, the model training module is used to train the first encoding model and the second encoding model based on the information noise contrast estimation loss function, wherein the information noise contrast estimation loss function is obtained by adjusting the sample similarity distribution.

[0116] The apparatus provided in this disclosure can achieve deep semantic parsing of unstructured data and improve the feature extraction capability of multi-source heterogeneous information by setting different encoding algorithms for structured fields and text description fields. Thus, this disclosure can fully consider the allocation of business data from multiple dimensions, improve information utilization, and enhance the reliability and business adaptability of data allocation.

[0117] It should be noted that the data acquisition module 601, business feature vector determination module 602, supplier feature vector determination module 603, similarity determination module 604, and target supplier determination module 605 mentioned above correspond to S202 to S210 in the method embodiment. The examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above method embodiment. It should be noted that the above modules, as part of the apparatus, can be executed in a computer system such as a set of computer-executable instructions.

[0118] Those skilled in the art will understand that various aspects of this disclosure can be implemented as a system, method, or program product. Therefore, various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0119] This disclosure provides an electronic device, which, exemplarily, includes a processor and a memory. The memory can be used to store executable instructions of the processor. The processor is configured to perform a supplier data allocation method provided in this disclosure by executing the executable instructions.

[0120] The following reference Figure 7 To describe an electronic device 700 according to such an embodiment of the present disclosure. Figure 7 The electronic device 700 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0121] like Figure 7 As shown, the electronic device 700 is presented in the form of a general-purpose computing device. The components of the electronic device 700 may include, but are not limited to: at least one processing unit 710, at least one storage unit 720, and a bus 730 connecting different system components (including the storage unit 720 and the processing unit 710).

[0122] The storage unit stores program code that can be executed by the processing unit 710, causing the processing unit 710 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of the present disclosure. For example, the processing unit 710 can perform the following steps of the above method embodiments: The process involves: acquiring business data and supplier data corresponding to multiple suppliers to be assigned; encoding the structured fields and text description fields in the business data using a first encoding model, and fusing the first structured feature and the first text feature obtained from the encoding to obtain a business feature vector; encoding the structured fields and text description fields in the supplier data using a second encoding model, and fusing the second structured feature and the second text feature obtained from the encoding to obtain a supplier feature vector corresponding to each supplier; determining multiple similarities based on the business feature vector and the multiple supplier feature vectors, where each business feature vector corresponds to one similarity with any supplier feature vector; and determining the target supplier from the multiple suppliers to be assigned based on the similarity.

[0123] Storage unit 720 may include readable media in the form of volatile storage units, such as random access memory (RAM) 7201 and / or cache memory 7202, and may further include read-only memory (ROM) 7203.

[0124] The storage unit 720 may also include a program / utility 7204 having a set (at least one) program module 7205, such program module 7205 including but not limited to: an operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0125] Bus 730 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0126] Electronic device 700 can also communicate with one or more external devices 740 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 700, and / or with any device that enables electronic device 700 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 750. Furthermore, electronic device 700 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 760. As shown, network adapter 760 communicates with other modules of electronic device 700 via bus 730. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 700, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0127] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0128] In particular, according to embodiments of this disclosure, the process described above with reference to the flowchart can be implemented as a computer program product, which includes a computer program that, when executed by a processor, implements the above-described supplier data allocation method.

[0129] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a computer program is stored. When executed by a processor, the computer program can implement the supplier data allocation method provided in the embodiments of this disclosure. The computer-readable storage medium may be a readable signal medium or a readable storage medium.

[0130] Figure 8 A schematic diagram of a computer-readable storage medium according to an embodiment of the present disclosure is shown, such as... Figure 8As shown, the computer-readable storage medium 800 stores a program product capable of implementing the methods described above in this disclosure. In some possible embodiments, various aspects of this disclosure may also be implemented as a program product comprising program code that, when run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure.

[0131] More specific examples of computer-readable storage media in this disclosure may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0132] In this disclosure, a computer-readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of transmitting, propagating, or transmitting a program for use by or in connection with an instruction execution system, apparatus, or device.

[0133] Optionally, the program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0134] In practical implementation, program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0135] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0136] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0137] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, mobile terminal, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0138] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope of this disclosure is indicated by the appended claims.

Claims

1. A method of distributing vendor data, characterized by, The method comprises: obtaining business data and supplier data corresponding to a plurality of to-be-assigned suppliers; encoding structured fields and text description fields in the business data through a first encoding model, and performing fusion processing on the encoded first structured features and first text features to obtain a business feature vector; encoding structured fields and text description fields in the supplier data through a second encoding model, and performing fusion processing on the encoded second structured features and second text features to obtain a supplier feature vector corresponding to each supplier; determining a plurality of similarities according to the business feature vector and the plurality of supplier feature vectors, wherein the business feature vector and any supplier feature vector correspond to one similarity; determining a target supplier from the plurality of to-be-assigned suppliers according to the similarities.

2. The provider data distribution method of claim 1, wherein, The method of determining a target supplier from the plurality of to-be-assigned suppliers according to the similarities comprises: determining N candidate suppliers from the plurality of to-be-assigned suppliers according to the similarities, N being a positive integer; generating a scoring result of each of the N candidate suppliers based on at least one of the similarities, supplier performance data describing historical performance information of the suppliers, and emergency degree matching data describing a matching degree between historical response times of the suppliers and a current business emergency degree; and determining at least one candidate supplier as the target supplier according to the scoring result.

3. The method of claim 1, wherein The method of encoding structured fields and text description fields in the business data through a first encoding model comprises: encoding the structured fields in the business data through a full connection layer in the first encoding model to obtain first structured features, and encoding the text description fields in the business data through a code encoder in the first encoding model to obtain first text features; wherein the fusion processing on the encoded first structured features and first text features to obtain a business feature vector comprises: fusion processing on the first structured features and the first text features based on a multi-head attention fusion algorithm to obtain a business feature vector.

4. The provider data distribution method of claim 1, wherein, The method of encoding structured fields and text description fields in the supplier data through a second encoding model comprises: encoding the text description fields in the supplier data through a text encoder in the second encoding model to obtain second structured features, and encoding the structured fields in the supplier data through a statistical encoder in the second encoding model to obtain second text features; wherein the fusion processing on the encoded second structured features and second text features to obtain a supplier feature vector corresponding to each supplier comprises: fusion processing on the second structured features and the second text features based on a dynamic gating network to obtain a supplier feature vector.

5. The provider data distribution method according to any one of claims 1 to 4, wherein, Before the method of obtaining business data and supplier data corresponding to a plurality of to-be-assigned suppliers, the method further comprises: Obtaining a positive sample set and a negative sample set, wherein the positive sample is a combination of a historical successful matched business and a supplier, and the negative sample is a combination of a business and a supplier that is not matched and / or not transacted; Training the first encoding model and the second encoding model through the positive sample set and the negative sample set.

6. The provider data distribution method of claim 5, wherein, The training of the first encoding model and the second encoding model through the positive sample set and the negative sample set comprises: The first encoding model and the second encoding model are trained in two stages, wherein the first stage training is the training of the target parameters in the first encoding model and the second encoding model after freezing, and the second stage training is the full parameter training of the first encoding model and the second encoding model.

7. The method of claim 5, wherein, The training of the first encoding model and the second encoding model through the positive sample set and the negative sample set comprises: According to the information noise contrast estimation loss function, the first encoding model and the second encoding model are trained, wherein the information noise contrast estimation loss function is adjusted based on the sample similarity distribution.

8. A provider data distribution apparatus, characterized by, Comprise: A data acquisition module for acquiring business data and supplier data corresponding to a plurality of to-be-allocated suppliers; A business feature vector determination module for encoding structured fields and text description fields in the business data through a first encoding model, and fusing first structured features and first text features obtained by encoding to obtain a business feature vector; A supplier feature vector determination module for encoding structured fields and text description fields in the supplier data through a second encoding model, and fusing second structured features and second text features obtained by encoding to obtain a supplier feature vector corresponding to each supplier; A similarity determination module for determining a plurality of similarities according to the business feature vector and a plurality of supplier feature vectors, wherein the business feature vector and any supplier feature vector correspond to one similarity; A target supplier determination module for determining a target supplier from a plurality of to-be-allocated suppliers according to the similarity.

9. An electronic device, comprising: Comprise: A processor; And A memory for storing executable instructions of the processor; Wherein the processor is configured to execute the executable instructions to perform the supplier data allocation method of any one of claims 1-7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the supplier data allocation method of any one of claims 1-7.

11. A computer program product, characterised in that, The computer program product comprises computer instructions stored in a computer readable storage medium, and the processor of the electronic device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the electronic device executes the supplier data allocation method as claimed in any one of claims 1-7.

Images