A method and device for classifying recruitment needs, an electronic device and a medium

Abstract

The application relates to a kind of classification methods, devices, electronic equipment and medium of recruitment needs, which include: the standard demand text corresponding to recruitment needs is cut and each recruitment needs corresponding tuple is obtained;The demand name and demand content in tuple are spliced to obtain spliced text, and the spliced text is added to tuple as element to obtain first tuple;Spliced text is input into the well-trained deep learning model, the vectorization representation corresponding to spliced text is output by the well-trained deep learning model, and the vectorization representation is added to first tuple as element to obtain second tuple;Based on hierarchical clustering algorithm, the set formed by the second tuple corresponding to recruitment needs is clustered, and the same recruitment needs are classified into a class.The application improves the efficiency of recruitment needs overall planning.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, and in particular to a method, apparatus, electronic device, and medium for classifying procurement needs. Background Technology

[0002] When a project is established, project management members need to submit procurement requirements from four professional lines: materials, equipment, labor subcontracting, and professional subcontracting, and identify suppliers. For example, it is necessary to clarify the supply channels for materials such as waterproof coatings, scaffolding, steel bars, and cement. The project management personnel are responsible for submitting specific requirements and quantities, and the relevant management personnel in the branch office coordinate the aggregated requirements from various projects. Requirements for projects that can be procured simultaneously are merged for procurement, i.e., procurement coordination, to achieve the goal of negotiating lower prices through increased volume. A project may submit dozens or even hundreds of procurement requirements, and a single branch office may be procuring for dozens of projects simultaneously. Therefore, branch office management personnel have a responsibility to coordinate and recommend suppliers.

[0003] In existing technologies, there are two main methods for coordinating procurement in supply chain systems: The first method involves project managers manually merging procurement requirements. This method requires a high level of professional skills from project managers and is prone to oversights. The second method uses a literal matching algorithm based on the textual descriptions of procurement requirements to merge them. This method relies excessively on literal matching and cannot match synonyms or near-synonyms in the procurement requirements. It only uses the textual part of the procurement requirements for matching and does not utilize the text containing richer information about the work content of the requirements, thus resulting in low efficiency. Summary of the Invention

[0004] In view of this, it is necessary to provide a method, device, electronic equipment and medium for classifying procurement needs, so as to improve the efficiency of overall planning of procurement needs.

[0005] To achieve the above objectives, the present invention provides a method for classifying procurement needs, comprising:

[0006] The standard requirement text corresponding to the procurement requirements is segmented into words to obtain the tuple corresponding to each procurement requirement.

[0007] The requirement name and requirement content in the tuple are concatenated to obtain concatenated text, and the concatenated text is added as an element to the tuple to obtain the first tuple;

[0008] The concatenated text is input into a fully trained deep learning model, and the fully trained deep learning model outputs the vectorized representation corresponding to the concatenated text. The vectorized representation is then added as an element to the first tuple to obtain the second tuple.

[0009] The hierarchical clustering algorithm is used to cluster the set of second tuples corresponding to the procurement requirements, and the same procurement requirements are grouped into one category.

[0010] In some possible implementations, the method further includes obtaining the fully trained deep learning model, wherein obtaining the fully trained deep learning model includes:

[0011] Establish a dataset of procurement demand samples, which includes a training set, a test set, and a validation set;

[0012] The training set is input into a pre-established initial neural network model for training to obtain a transitional neural network model;

[0013] The transitional neural network model is validated based on the validation set, and it is determined whether the validated model has reached the prediction accuracy or the number of iterations. If it has not reached the required accuracy or the required number of iterations, the transitional neural network model is validated again. If it has reached the required accuracy or the required number of iterations, the transitional neural network model is determined to be the target neural network model.

[0014] The predictive performance of the target neural network model is tested using the test set to obtain a fully trained deep learning model.

[0015] In some possible implementations, the deep learning model is the BERT model.

[0016] In some possible implementations, the standard requirement text includes: requirement name, requirement content, requirement ID, and requirement time.

[0017] In some possible implementations, the hierarchical clustering algorithm is used to cluster the set of second tuples corresponding to the procurement requirements, grouping the same procurement requirements into one category, including:

[0018] Clustering is performed based on the pairwise cosine distance between the vectorized representations of each element in the set, grouping those with the same procurement needs into one category.

[0019] In some possible implementations, clustering based on the pairwise cosine distance between the vectorized representations of each element in the set, grouping similar procurement needs into one category, includes:

[0020] Calculate the pairwise cosine distance between the vectorized representations of each element in the set;

[0021] The two clusters whose pairwise cosine distance is less than a preset threshold are merged and repeated until the preset threshold is met to obtain several clusters. Each element in the set corresponds to one of the several clusters. The main cluster corresponding to each element in the set is added as an element to the second tuple to obtain the third tuple.

[0022] The procurement requirements corresponding to each cluster are sorted according to the required time in the procurement requirements of each cluster.

[0023] In each of the several clusters, clusters whose demand time for adjacent procurement needs differs by more than a preset time are divided into sub-clusters. The sub-cluster corresponding to each element in the set is added as an element to the third tuple to obtain the fourth tuple.

[0024] Aggregate the procurement needs with the same main cluster and sub-cluster values ​​in the fourth tuple, and group the procurement needs with the same values ​​into one category based on the aggregation result.

[0025] In some possible implementations, the formula for calculating the pairwise cosine distance between the vectorized representations of each element in the set is:

[0026]

[0027] In the formula, distance(C) i C j ) represents C in the set. i Clusters and C j The pairwise cosine distance between the vectorized representations of clusters; p represents C i Any element in the cluster; p ′ Indicate C j Any element in a cluster; embedding p Represents the vectorized representation of p; embedding p ′ represents p ′ Vectorized representation of.

[0028] On the other hand, the present invention also provides a classification device for procurement needs, comprising:

[0029] The tuple acquisition module is used to extract the standard requirement text corresponding to the procurement requirements and perform word segmentation to obtain the tuple corresponding to each procurement requirement.

[0030] The first tuple acquisition module is used to concatenate the requirement name and requirement content in the tuple to obtain concatenated text, and add the concatenated text as an element to the tuple to obtain the first tuple.

[0031] The second tuple acquisition module is used to input the concatenated text into a fully trained deep learning model, output the vectorized representation corresponding to the concatenated text through the fully trained deep learning model, and add the vectorized representation as an element to the first tuple to obtain the second tuple.

[0032] The procurement demand classification module is used to cluster the set of second tuples corresponding to the procurement demand based on the hierarchical clustering algorithm, and group the same procurement demand into one category.

[0033] On the other hand, the present invention also provides an electronic device, including a memory and a processor, wherein,

[0034] The memory is used to store programs;

[0035] The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the procurement demand classification method described in any of the above implementations.

[0036] On the other hand, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps in the procurement demand classification method described in any of the above implementations.

[0037] The beneficial effects of the above embodiments are as follows: The method for classifying procurement needs provided by the present invention first segments the standard requirement text corresponding to the procurement needs into tuples for each procurement need. The requirement name and requirement content are then concatenated to obtain concatenated text. This concatenated text is added as an element to the tuples to obtain the first tuple. The concatenated text is then input into a fully trained deep learning model to obtain a vectorized representation of the concatenated text. This vectorized representation is added to the first tuple to obtain the second tuple. Finally, a hierarchical clustering algorithm is used to cluster the set of the second tuples corresponding to the procurement needs, grouping similar procurement needs into one category. The present invention uses the procurement need text and the procurement need work content as input, inputs them into a deep learning model to obtain vectorized representations, and finally uses a hierarchical clustering algorithm to group similar procurement needs into one category, thus improving the efficiency of procurement need coordination. Attached Figure Description

[0038] Figure 1 A flowchart illustrating an embodiment of a method for classifying procurement needs provided by the present invention;

[0039] Figure 2 A schematic diagram of an embodiment of a procurement demand classification device provided by the present invention;

[0040] Figure 3 A schematic diagram of an embodiment of the electronic device provided by the present invention. Detailed Implementation

[0041] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0042] Figure 1 A schematic flowchart of an embodiment of a method for classifying procurement needs provided by the present invention is shown below. Figure 1 As shown, a method for classifying procurement needs includes:

[0043] S101. Take the standard requirement text corresponding to the procurement requirement and perform word segmentation to obtain the tuple corresponding to each procurement requirement;

[0044] S102. Concatenate the requirement name and requirement content in the tuple to obtain concatenated text, and add the concatenated text as an element to the tuple to obtain the first tuple.

[0045] S103. Input the concatenated text into a fully trained deep learning model, output the vectorized representation corresponding to the concatenated text through the fully trained deep learning model, and add the vectorized representation as an element to the first tuple to obtain the second tuple.

[0046] S104. Based on the hierarchical clustering algorithm, cluster the set of second tuples corresponding to the procurement requirements, and group the same procurement requirements into one category.

[0047] Compared with existing technologies, this embodiment provides a method for classifying procurement needs. First, the standard requirement text corresponding to each procurement need is segmented to obtain tuples for each requirement. The requirement name and content are then concatenated to obtain concatenated text. This concatenated text is added as an element to the tuples to obtain the first tuple. Next, the concatenated text is input into a fully trained deep learning model to obtain a vectorized representation of the concatenated text. This vectorized representation is then added to the first tuple to obtain the second tuple. Finally, a hierarchical clustering algorithm is used to cluster the set of second tuples corresponding to the procurement needs, grouping similar procurement needs into the same category. This invention uses the procurement need text and the procurement need work content as input, inputs them into a deep learning model to obtain vectorized representations, and finally uses a hierarchical clustering algorithm to group similar procurement needs into the same category, thus improving the efficiency of procurement need coordination.

[0048] In a specific embodiment of the present invention, step S101 involves organizing the procurement requirements into standard requirement text and performing word segmentation to obtain tuples corresponding to each procurement requirement, including:

[0049] Collect all project procurement requirements, the specific work content of each requirement, the unique ID of each procurement requirement, and the entry time of each supplier for each procurement requirement. Each procurement requirement can be represented as a tuple (project_name, project_content, project_id, project_time).

[0050] In a specific embodiment of the present invention, in step S102, the requirement name and requirement content in the tuple are concatenated to obtain concatenated text, and the concatenated text is added as an element to the tuple to obtain a first tuple, including:

[0051] The `project_name` and `project_content` are concatenated as input for this procurement requirement. For example, if `project_name` is "Doors and Windows Project" and `project_content` is "Fireproof Doors and Windows Project", we use "[SEP]" to concatenate `project_name` and `project_content`, and insert "[CLS]" at the beginning. The preprocessed text result for this procurement requirement is `[CLS]Doors and Windows Project [SEP]Fireproof Doors and Windows Project`. The preprocessed result is denoted as `project_text`. Each procurement requirement can be represented as (project_name, project_content, project_id, project_time, project_text), i.e., the first tuple.

[0052] In some embodiments of the present invention, it further includes:

[0053] Establish a dataset of procurement demand samples, which includes a training set, a test set, and a validation set;

[0054] The training set is input into a pre-established initial neural network model for training to obtain a transitional neural network model;

[0055] The transitional neural network model is validated based on the validation set, and it is determined whether the validated model has reached the prediction accuracy or the number of iterations. If it has not reached the required accuracy or the required number of iterations, the transitional neural network model is validated again. If it has reached the required accuracy or the required number of iterations, the transitional neural network model is determined to be the target neural network model.

[0056] The predictive performance of the target neural network model is tested using the test set to obtain a fully trained deep learning model.

[0057] In some embodiments of the present invention, the deep learning model is the BERT model.

[0058] In a specific embodiment of the present invention, in step S103, the project_text from the unary tuple obtained in step S102 is input into the deep learning model BERT, and the vector corresponding to [CLS] in the output of BERT is used as the output result. At this time, each procurement requirement can be represented as (project_name, project_content, project_id, project_time, project_text, embedding).

[0059] The BERT model was fine-tuned using a contrastive learning approach.

[0060] Sample selection: The historical data of procurement requirements that were subject to unified procurement are merged, resulting in N sets. For each set, 10 procurement requirements that are not in the set are randomly selected from the historical procurement requirements as negative samples, and each procurement requirement in the set is used as a positive sample. Each set of positive and negative samples is used as a training data point, and the positive and negative samples are processed as described in step 1.

[0061] Model training: The training data obtained from sample selection is input into the BERT model, and the output vector is obtained through inference. Assuming there are k samples in a training dataset, where positive samples are denoted as k+, the training objective function L can be written as: Here, τ is the temperature coefficient, which we set to 0.05. The deep learning model parameters are updated using gradient descent and backpropagation to achieve the desired model training effect.

[0062] In some embodiments of the present invention, the standard requirement text includes: requirement name, requirement content, requirement ID, and requirement time.

[0063] In some embodiments of the present invention, the clustering of the set of second tuples corresponding to the procurement requirements based on the hierarchical clustering algorithm, and grouping the same procurement requirements into one category, includes:

[0064] Clustering is performed based on the pairwise cosine distance between the vectorized representations of each element in the set, grouping those with the same procurement needs into one category.

[0065] In some embodiments of the present invention, the step of clustering based on the pairwise cosine distance between the vectorized representations of each element in the set to group similar procurement needs into one category includes:

[0066] Calculate the pairwise cosine distance between the vectorized representations of each element in the set;

[0067] Merge two clusters with pairwise cosine distances less than a preset threshold, and repeat continuously until the preset threshold is satisfied, obtaining several clusters. Each element in the set corresponds to one of the several clusters. Add the main cluster corresponding to each element in the set as an element to the second tuple to obtain a third tuple;

[0068] Sort the procurement requirements corresponding to each cluster in the several clusters according to the required time in the procurement requirements;

[0069] In each of the several clusters, split the clusters where the difference in required time between adjacent procurement requirements is greater than the preset time to obtain sub-clusters respectively. Add the sub-clusters corresponding to each element in the set as elements to the third tuple to obtain a fourth tuple;

[0070] Aggregate the procurement requirements with the same main cluster and sub-cluster values in the fourth tuple, and classify the same procurement requirements according to the aggregation result.

[0071] In a specific embodiment of the present invention, the set of all procurement requirements is denoted as D, and hierarchical clustering is used to cluster according to the pairwise cosine distances of the embeddings of the procurement requirements. The steps are as follows:

[0072] Each element in the set D is a cluster, and calculate the distance between two clusters in the set D.

[0073] In some embodiments of the present invention, the calculation formula for the pairwise cosine distances between the vectorized representations of each element in the set is:

[0074]

[0075] In the formula, distance(C i , C j ) represents the pairwise cosine distance between the vectorized representations between the C i cluster and the C j cluster in the set; p represents any element in the C i cluster; p' represents any element in the C j cluster; embedding p represents the vectorized representation of p; embedding p′ represents the vectorized representation of p'.

[0076] Determine two clusters C i , C j with the smallest distance between two clusters in the set D. If distance(C i , C j ) < threshold, where threshold is a preset value, then C iC j Merge them into a new cluster; thus obtaining N clusters. For each cluster, given the cluster number 1-N, each procurement requirement will belong to a unique cluster. Each procurement requirement can be represented as (project_name, project_content, project_id, project_time, project_text, embedding, cluster_class).

[0077] Within each cluster, procurement requirements are sorted in ascending order of supplier entry time (project_time). If adjacent procurement requirements have a project_time difference exceeding 60 days, the cluster is split into two sub-clusters, and further post-processing is performed on the latter sub-cluster. A sub-cluster is denoted as `sub_class`, and each procurement requirement can be represented as (project_name, project_content, project_id, project_time, project_text, embedding, cluster_class, sub_class). Procurement requirements with the same `cluster_class` and `sub_class` are aggregated together, grouping identical requirements into one category.

[0078] To better implement the procurement demand classification method in this embodiment of the invention, based on the procurement demand classification method, correspondingly, as follows: Figure 2 As shown, this embodiment of the invention also provides a device for classifying procurement needs. A procurement needs classification device 200 includes:

[0079] The tuple acquisition module 201 is used to extract the standard requirement text corresponding to the procurement requirement and perform word segmentation to obtain the tuple corresponding to each procurement requirement.

[0080] The first tuple acquisition module 202 is used to concatenate the requirement name and requirement content in the tuple to obtain concatenated text, and add the concatenated text as an element to the tuple to obtain the first tuple.

[0081] The second tuple acquisition module 203 is used to input the concatenated text into a fully trained deep learning model, output the vectorized representation corresponding to the concatenated text through the fully trained deep learning model, and add the vectorized representation as an element to the first tuple to obtain the second tuple.

[0082] The procurement demand classification module 204 is used to cluster the set of second tuples corresponding to the procurement demand based on the hierarchical clustering algorithm, and group the same procurement demand into one category.

[0083] The procurement demand classification device 200 provided in the above embodiment can realize the technical solution described in the above embodiment of the procurement demand classification method. The specific implementation principle of each module or unit can be found in the corresponding content in the above embodiment of the procurement demand classification method, which will not be repeated here.

[0084] like Figure 3 As shown, the present invention also provides an electronic device 300. The electronic device 300 includes a processor 301, a memory 302, and a display 303. Figure 3 Only some components of the electronic device 300 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0085] In some embodiments, processor 301 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 302 or process data, such as a method for classifying procurement requirements in this invention.

[0086] In some embodiments, processor 301 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 301 may be local or remote. In some embodiments, processor 301 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, intranet, multi-cloud, etc., or any combination thereof.

[0087] In some embodiments, memory 302 may be an internal storage unit of electronic device 300, such as a hard disk or memory of electronic device 300. In other embodiments, memory 302 may also be an external storage device of electronic device 300, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. equipped on electronic device 300.

[0088] Furthermore, the memory 303 may include both internal storage units of the electronic device 300 and external storage devices. The memory 302 is used to store application software and various types of data installed on the electronic device 300.

[0089] In some embodiments, display 303 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 303 is used to display information from electronic device 300 and to display a visual user interface. Components 301-303 of electronic device 300 communicate with each other via a system bus.

[0090] In one embodiment, when processor 301 executes a classification program for procurement needs stored in memory 302, the following steps can be implemented:

[0091] The standard requirement text corresponding to the procurement requirements is segmented into words to obtain the tuple corresponding to each procurement requirement.

[0092] The requirement name and requirement content in the tuple are concatenated to obtain concatenated text, and the concatenated text is added as an element to the tuple to obtain the first tuple;

[0093] The concatenated text is input into a fully trained deep learning model, and the fully trained deep learning model outputs the vectorized representation corresponding to the concatenated text. The vectorized representation is then added as an element to the first tuple to obtain the second tuple.

[0094] The hierarchical clustering algorithm is used to cluster the set of second tuples corresponding to the procurement requirements, and the same procurement requirements are grouped into one category.

[0095] It should be understood that when the processor 301 executes a procurement demand classification program in the memory 302, in addition to the functions mentioned above, it can also perform other functions, as detailed in the description of the corresponding method embodiments above.

[0096] Furthermore, the embodiments of the present invention do not specifically limit the type of electronic device 300 mentioned. Electronic device 300 can be a mobile phone, tablet computer, personal digital assistant (PDA), wearable device, laptop computer, or other portable electronic device. Exemplary embodiments of portable electronic devices include, but are not limited to, portable electronic devices running iOS, Android, Microsoft, or other operating systems. The aforementioned portable electronic device can also be other portable electronic devices, such as a laptop computer with a touch-sensitive surface (e.g., a touch panel). It should also be understood that in some other embodiments of the present invention, electronic device 300 may not be a portable electronic device, but rather a desktop computer with a touch-sensitive surface (e.g., a touch panel).

[0097] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0098] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method of classifying a demand for recruitment, characterized by, The method comprises the following steps: corresponding to the standard demand text of the bidding demand and performing word segmentation to obtain a tuple corresponding to each bidding demand; splicing the demand name and the demand content in the tuple to obtain a spliced text, and adding the spliced text as an element to the tuple to obtain a first tuple; inputting the spliced text into a trained deep learning model, outputting a vectorized representation corresponding to the spliced text through the trained deep learning model, and adding the vectorized representation as an element to the first tuple to obtain a second tuple; based on a hierarchical clustering algorithm, clustering a set composed of the second tuples corresponding to the bidding demands, and grouping the same bidding demands into a category; according to the cosine distance between the vectorized representations of each element in the set, clustering, and grouping the same bidding demands into a category, comprising: calculating the cosine distance between the vectorized representations of each element in the set; merging two clusters with a cosine distance less than a preset threshold, and repeating until the preset threshold is met, to obtain a plurality of clusters, each element in the set corresponding to one of the plurality of clusters, and adding the main cluster corresponding to each element in the set as an element to the second tuple to obtain a third tuple; sorting the bidding demands corresponding to each cluster in the plurality of clusters according to the demand time in the bidding demands corresponding to each cluster; in each cluster in the plurality of clusters, cutting the cluster in which the demand time of adjacent bidding demands differs by more than a preset time to obtain sub-clusters, and adding the sub-cluster corresponding to each element in the set as an element to the third tuple to obtain a fourth tuple; aggregating the bidding demands with the same main cluster and sub-cluster value in the fourth tuple, and grouping the same bidding demands into a category according to the aggregation result; the formula for calculating the cosine distance between the vectorized representations of each element in the set is: wherein, denotes the cosine distance between the vectorized representation of any element in the set cluster and the vectorized representation of any element in the set denotes any element in the set denotes any element in the set denotes the vectorized representation of p denotes the vectorized representation of 2. The method of claim 1, wherein, further comprising: establishing a data set of bidding demand samples, wherein the bidding demand sample data set comprises a training set, a test set and a validation set; inputting the training set into a pre-established initial neural network model for training to obtain a transition neural network model; based on the validation set, verifying the transition neural network model and determining whether the verified model reaches the prediction accuracy or the number of iterations, if not, verifying the transition neural network model again, if so, determining that the transition neural network model is a target neural network model; testing the prediction performance of the target neural network model using the test set to obtain a trained deep learning model.

3. The method of claim 2, wherein the method further comprises: The deep learning model is a BERT model.

4. The method of claim 1, wherein the method further comprises: The standard demand text includes demand name, demand content, demand id and demand time.

5. The method of claim 1, wherein the method further comprises: based on the hierarchical clustering algorithm, clustering the set composed of the second tuples corresponding to the bidding demands, and grouping the same bidding demands into a category, comprising: according to the cosine distance between the vectorized representations of each element in the set, clustering, and grouping the same bidding demands into a category.

6. A device for classifying recruitment needs, characterized in that comprising: a tuple acquisition module configured to obtain a tuple corresponding to each bidding demand by performing word segmentation on the standard demand text corresponding to the bidding demand; The first tuple obtaining module is configured to splice the demand name and the demand content in the tuple to obtain a spliced text, and add the spliced text as an element into the tuple to obtain a first tuple; The second tuple obtaining module is configured to input the spliced text into a trained deep learning model, output a vectorized representation corresponding to the spliced text through the trained deep learning model, and add the vectorized representation as an element into the first tuple to obtain a second tuple; The bidding and purchasing demand classification module is configured to cluster a set composed of the second tuples corresponding to the bidding and purchasing demands based on a hierarchical clustering algorithm, and classify the same bidding and purchasing demands into a category; The bidding and purchasing demand classification module is configured to cluster the set based on a cosine distance between vectorized representations of each element in the set, and classify the same bidding and purchasing demands into a category, including: The bidding and purchasing demand classification module is configured to calculate the cosine distance between vectorized representations of each element in the set; The bidding and purchasing demand classification module is configured to merge two clusters with a cosine distance less than a preset threshold, and repeatedly until the preset threshold is met, to obtain a plurality of clusters, and add a main cluster corresponding to each element in the set as an element into the second tuple to obtain a third tuple; The bidding and purchasing demand classification module is configured to sort the bidding and purchasing demands corresponding to each cluster in the plurality of clusters according to a demand time in the bidding and purchasing demands corresponding to each cluster; The bidding and purchasing demand classification module is configured to cut the cluster in which the demand times of adjacent bidding and purchasing demands differ by more than a preset time to obtain sub-clusters, and add a sub-cluster corresponding to each element in the set as an element into the third tuple to obtain a fourth tuple; The bidding and purchasing demand classification module is configured to aggregate the bidding and purchasing demands with the same main cluster and sub-cluster value in the fourth tuple, and classify the same bidding and purchasing demands into a category according to the aggregation result. The cosine distance between vectorized representations of each element in the set is calculated according to the following formula: wherein, denotes the cosine distance between the vectorized representation of any element in the set cluster and the vectorized representation of any element in the set denotes the cosine distance between the vectorized representation of any element in the set cluster and the vectorized representation of any element in the set cluster and denotes the vectorized representation of p denotes the vectorized representation of the vectorized representation of 7. An electronic device, comprising: The memory is configured to store a program; The processor is coupled to the memory and is configured to execute the program stored in the memory to implement the steps of the method for classifying bidding and purchasing demands according to any one of claims 1 to 5. The memory is configured to store a program or instructions readable by a computer, which are executed by a processor to implement the steps of the method for classifying bidding and purchasing demands according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, ​

Images