Model training method, recommendation method, device, equipment, medium and program product

Abstract

The invention provides a model training method and device, a recommendation method and device, equipment, a medium and a program product, and relates to the technical field of computers. The recommendation model training method comprises the steps that feature extraction is conducted on historical process data of a process management system, text features, statistical features and network features are obtained, the text features are used for representing description information of a process, the statistical features are used for representing processing information of the process, and the network features are used for representing network information of the process; the network features are used for representing association information among historical processing persons of the processing flow; constructing a to-be-trained model based on the pre-training language model and the deep learning model; and performing model training on the to-be-trained model based on the text features, the statistical features and the network features to obtain a recommendation model, and recommending a processor to the to-be-processed process by the recommendation model. Through the technical scheme disclosed by the invention, the association relationship between the process link and the processing person is found by mining and analyzing the historical process data, and the reliable process processing person is recommended to optimize the process management process.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to a recommendation model training method, a process handler recommendation method, a recommendation model training device, a process handler recommendation device, an electronic device, a computer-readable storage medium, and a computer program product. Background Art

[0002] The process management system takes business processes as its core, and helps enterprises or organizations improve operational efficiency by modeling, automating and continuously monitoring processes. In a process management system, the person who handles a process is usually manually specified or relies on preset designated rules. This method not only requires a lot of manpower investment, but also appears to be inflexible when facing complex and changing business scenarios. Especially for processes that do not have preset handlers, users often need to find suitable handlers through offline communication and other methods, which not only increases communication costs, but also reduces the efficiency of process processing.

[0003] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0004] The purpose of the present disclosure is to provide a recommendation model training method, a process handler recommendation method, a recommendation model training device, a process handler recommendation device, an electronic device, a computer-readable storage medium and a computer program product, which at least to a certain extent overcome the problem of low efficiency of process processing in process management systems in related technologies.

[0005] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0006] According to one aspect of the present disclosure, a method for recommending process handlers is provided, comprising: extracting features from historical process data of a process management system to obtain text features, statistical features, and network features, wherein the text features are used to characterize description information of the process, the statistical features are used to characterize processing information of the process, and the network features are used to characterize association information between historical handlers of the processing process; constructing a model to be trained based on a pre-trained language model and a deep learning model; and training the model to be trained based on the text features, the statistical features, and the network features to obtain the recommendation model, so that the recommendation model recommends handlers for the process to be processed.

[0007] In one embodiment of the present disclosure, a model to be trained is constructed based on a pre-trained language model and a deep learning model, including: inputting the historical process data marked with recommended handlers into the pre-trained language model, and calculating the difference between the recommendation result and the recommended handler; adjusting the model parameters of the pre-trained language model based on the difference to obtain a fine-tuned language model suitable for process management; and constructing the model to be trained based on the fine-tuned language model and the deep learning model.

[0008] In one embodiment of the present disclosure, constructing the model to be trained based on the fine-tuning language model and the deep learning model includes: using the fine-tuning language model as a front-end model and using the deep learning model as a back-end model to construct the model to be trained.

[0009] In one embodiment of the present disclosure, the model to be trained is trained based on the text features, the statistical features and the network features to obtain the recommendation model, including: selecting a corresponding loss function and optimizer according to the task type of the recommendation handler; inputting the text features, the statistical features and the network features into the fine-tuning language model to output a fused feature vector; inputting the fused feature vector into the deep learning model to capture the timing features and dependencies between different historical processes based on the deep learning model, and outputting a recommendation label; calculating the loss between the recommendation label and the actual label of the recommendation handler based on the loss function; and optimizing the model parameters of the deep learning model based on the optimizer and the overfitting prevention strategy to obtain an optimized model; verifying the performance of the optimized model based on a validation set to adjust the hyperparameters of the optimized model to obtain the recommendation model.

[0010] In one embodiment of the present disclosure, constructing the model to be trained based on the fine-tuned language model and the deep learning model includes: respectively using the fine-tuned language model and the deep learning model as base learners in a tree model to construct the model to be trained.

[0011] In one embodiment of the present disclosure, the model to be trained is trained based on the text features, the statistical features and the network features to obtain the recommendation model, including: inputting the text features, the statistical features and the network features into the fine-tuning language model and the deep learning model to obtain a first output feature and a second output feature, respectively; concatenating the first output feature and the second output feature to obtain a concatenated feature; the tree model includes a decision tree, and the splitting features and splitting points of the decision tree are selected based on the concatenated feature; and cross-validating the tree model that generates the decision tree to adjust the parameters of the tree model based on the validation results to obtain the recommendation model.

[0012] In one embodiment of the present disclosure, feature extraction is performed on historical process data of a process management system to obtain text features, statistical features and network features, including: collecting the historical process data from the process management system, the historical process data including process type, process handler, processing information and supplementary information; preprocessing the historical process data to obtain normalized data; performing word segmentation and word embedding processing on the text information in the normalized data based on a BERT model to obtain the text features; counting the attribute information of the processing information to obtain the statistical features; and extracting the association and influence features between different process handlers in the organizational chart based on a graph neural network as the network features.

[0013] According to another aspect of the present disclosure, a method for recommending process handlers is provided, comprising: extracting process information in a generated process to be processed in response to the process to be processed; inputting the process information into a recommendation model for process handlers to output a recommended handler for processing the process to be processed and a corresponding recommendation probability, wherein the recommendation model is constructed based on a pre-trained language model and a deep learning model; and obtaining status information of the recommended handler to adjust the recommended handler and the corresponding recommendation probability based on the status information to generate a recommendation target.

[0014] In one embodiment of the present disclosure, in response to a generated process to be processed, process information in the process to be processed is extracted, including: in response to a generated process to be processed, process description information in the process to be processed, processing information on the process, and at least one of the association information between historical processors of the process to be processed are extracted as the process information.

[0015] In one embodiment of the present disclosure, the recommendation model is generated based on a fine-tuned language model and a deep learning model connected in series, the fine-tuned language model is generated based on fine-tuning the pre-trained language model, and the process information is input into the recommendation model of the process handler to output the recommended handler for processing the process to be processed and the corresponding recommendation probability, including: inputting the process information into the fine-tuned language model, outputting a fused feature vector describing the process to be processed, the fused feature vector fused with text features, statistical features and network features; inputting the fused feature vector into the deep learning model to capture the dependency relationship between the process to be processed and the completed process based on the deep learning model, and outputting the recommended handler for the process to be processed and the corresponding recommendation probability.

[0016] In one embodiment of the present disclosure, the recommendation model is generated based on respectively using the fine-tuned language model and the deep learning model as base learners in a tree model, the fine-tuned language model is generated based on fine-tuning the pre-trained language model, and the process information is input into the recommendation model of the process handler to output the recommended handler for processing the process to be processed and the corresponding recommendation probability, including: inputting the process information into the fine-tuned language model and the deep learning model to obtain a first output feature and a second output feature respectively; concatenating the first output feature and the second output feature to obtain a concatenated feature; inputting the concatenated feature into a decision tree in the tree model to output the recommended handler for the process to be processed and the corresponding recommendation probability.

[0017] In one embodiment of the present disclosure, it also includes: generating a display interface for the recommended target; in response to a selection operation on the interface, if the recommended target is selected, pushing the process to be processed to the terminal of the recommended target; if the recommended target is not selected, collecting the processor information specified by the user, and feeding back the specified processor information to the recommendation model for model optimization.

[0018] According to another aspect of the present disclosure, a recommendation model training device is provided, including: a feature extraction module, used to extract features from historical process data of a process management system to obtain text features, statistical features and network features, wherein the text features are used to characterize description information of the process, the statistical features are used to characterize processing information of the process, and the network features are used to characterize association information between historical processors of the processing process; a model construction module, used to construct a model to be trained based on a pre-trained language model and a deep learning model; and a model training module, used to perform model training on the model to be trained based on the text features, the statistical features and the network features to obtain the recommendation model, so that the recommendation model recommends processors for the process to be processed.

[0019] According to another aspect of the present disclosure, there is provided a device for recommending process handlers, comprising: an information extraction module, for extracting process information in the process to be processed in response to a generated process to be processed; a model processing module, for inputting the process information into a recommendation model for process handlers, so as to output a recommended handler for processing the process to be processed and a corresponding recommendation probability, wherein the recommendation model is constructed based on a pre-trained language model and a deep learning model; and an adjustment module, for obtaining status information of the recommended handler, so as to adjust the recommended handler and the corresponding recommendation probability based on the status information, and generate a recommendation target.

[0020] According to another aspect of the present disclosure, there is provided an electronic device, comprising: a processor; and a memory for storing executable instructions of the processor; the processor is configured to execute the above-mentioned method for recommending a process handler by executing the executable instructions.

[0021] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for recommending a process handler is implemented.

[0022] According to another aspect of the present disclosure, a computer program product is provided, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned method for recommending a process handler is implemented.

[0023] The recommended model training and application scheme for process handlers provided by the embodiments of the present disclosure parses text features, statistical features and network features from historical process data, and uses comprehensive multi-dimensional information to train the model to be trained. The model to be trained is constructed based on a pre-trained language model and a deep learning model. The pre-trained language model and the deep learning model are constructed by introducing an artificial intelligence algorithm. The pre-trained language model can capture rich semantic information in the process description, and the deep learning model can capture the dependencies and timing characteristics between process links, thereby realizing the mining and precise analysis of historical process data to discover the association between process links and handlers. The recommendation model for process handlers obtained based on model training can accurately identify personnel suitable for handling specific processes to ensure the reliability of the process handler recommendation and prevent the process from being assigned to handlers who may cause process bottlenecks, thereby improving the efficiency of process processing.

[0024] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0026] Figure 1 A schematic diagram showing a process processing person recommendation system according to an embodiment of the present disclosure;

[0027] Figure 2 A schematic flow chart showing a recommendation model training method in an embodiment of the present disclosure;

[0028] Figure 3 A schematic flow chart showing another recommendation model training method in an embodiment of the present disclosure;

[0029] Figure 4 A schematic flow chart showing another recommendation model training method in an embodiment of the present disclosure;

[0030] Figure 5 A schematic flow chart showing another recommendation model training method in an embodiment of the present disclosure;

[0031] Figure 6 A schematic flow chart showing a method for recommending a process handler in an embodiment of the present disclosure;

[0032] Figure 7 A schematic flow chart showing another method for recommending a process handler in an embodiment of the present disclosure;

[0033] Figure 8 A schematic diagram of a recommendation model training device in an embodiment of the present disclosure is shown;

[0034] Fig. 9 A schematic diagram of a device for recommending a process handler in an embodiment of the present disclosure is shown;

[0035] Fig.10 A structural block diagram of an electronic device in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0036] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example 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.

[0037] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0038] Business Process Management System (BPMS) is a software system used to design, execute, monitor and optimize business processes. In a business process management system, process processors are individuals or roles responsible for executing tasks in each link of the process. They undertake different tasks at different stages of the process, including but not limited to information input, business operations, approval decisions, coordination and communication, etc. They are key factors in promoting the smooth progress of the process. Taking the procurement process management system as an example, the processors of each link include the names and departments of the persons in charge of application, approval, procurement, acceptance, etc.

[0039] The current process management system lacks an intelligent recommendation mechanism, which leads to problems such as uncertainty in process handlers, low circulation efficiency and increased communication costs. In this disclosure, by introducing AI / big models and other models, historical process data is automatically analyzed to identify the association rules between each link and the handler, and the most suitable handler is recommended in real time when the new process is executed, thereby optimizing the process management process and improving process processing efficiency and accuracy.

[0040] Figure 1 1 is a schematic diagram of a computer system provided by an exemplary embodiment of the present application. The system includes: a plurality of terminals 120 and a server 140.

[0041] The terminal 120 may be a mobile terminal such as a mobile phone, a game console, a tablet computer, an e-book reader, smart glasses, an MP4 (Moving Picture Experts Group Audio Layer IV) player, a smart home device, an AR (Augmented Reality) device, a VR (Virtual Reality) device, or the terminal 120 may also be a personal computer (PC), such as a laptop computer and a desktop computer.

[0042] The terminal 120 may be installed with an application for providing recommendations for process handlers.

[0043] The terminal 120 is connected to the server 140 via a communication network. Optionally, the communication network is a wired network or a wireless network.

[0044] The server 140 is a server, or is composed of several servers, or is a virtualization platform, or is a cloud computing service center. The server 140 is used to provide background services for the application program that provides the recommendation of the process handler. Optionally, the server 140 undertakes the main computing work, and the terminal 120 undertakes the secondary computing work; or, the server 140 undertakes the secondary computing work, and the terminal 120 undertakes the main computing work; or, the terminal 120 and the server 140 adopt a distributed computing architecture for collaborative computing.

[0045] In some optional embodiments, the server 140 is used to store the recommended model training program of the process handler.

[0046] Those skilled in the art will appreciate that the number of the terminals 120 may be more or less. For example, there may be only one terminal, or there may be dozens or hundreds of terminals, or a greater number. The embodiment of the present application does not limit the number and device type of the terminals.

[0047] Optionally, the system may also include a management device ( Figure 1 (not shown), the management device is connected to the server 140 via a communication network. Optionally, the communication network is a wired network or a wireless network.

[0048] Optionally, the above-mentioned wireless network or wired network uses standard communication technology and / or protocol. The network is usually the Internet, but it can also be any network, including but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a dedicated network or any combination of a virtual private network). In some embodiments, the data exchanged through the network is represented by technologies and / or formats including Hyper Text Mark-up Language (HTML), Extensible Markup Language (XML), etc. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPsec) can also be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can also be used to replace or supplement the above-mentioned data communication technologies.

[0049] like Figure 2 As shown, a method for recommending a process handler according to an embodiment of the present disclosure includes:

[0050] Step S202, feature extraction is performed on the historical process data of the process management system to obtain text features, statistical features and network features. The text features are used to characterize the description information of the process, the statistical features are used to characterize the processing information of the process, and the network features are used to characterize the association information between the historical processors of the process.

[0051] like Figure 3 As shown, in one embodiment of the present disclosure, a specific implementation method of extracting features from historical process data of a process management system to obtain text features, statistical features, and network features includes:

[0052] Step S302, collecting historical process data from the process management system, the historical process data including process type, process processor, processing information and supplementary information.

[0053] In some embodiments, taking a procurement process management system as an example, the system records procurement process data for many years, and historical process data is derived from the system, including but not limited to:

[0054] Process type: including "raw material procurement" and "office supplies procurement" etc.

[0055] Process links and handler information: The process links include application, approval, procurement, acceptance, etc., as well as the name and department of the person in charge of each link.

[0056] Processing information includes processing time (the start and end time of each link) and processing results (the processing results of each link, including but not limited to approval, rejection and modification).

[0057] Supplementary information: including but not limited to process name, process description, handling opinions, notes and other text information.

[0058] Step S304, pre-processing the historical process data to obtain normalized data.

[0059] Obtain process-related text information from historical process data, including process name, detailed description, document content, approval opinions, etc. Use natural language processing technology, such as word segmentation and stop word removal, to pre-process these texts, including cleaning, deduplication and normalization, to obtain standardized data.

[0060] Among them, standardization refers to unifying the format of fields such as time and date to ensure the consistency and comparability of data.

[0061] In some embodiments, preprocessing historical process data also includes: using natural language processing technology (NLP) to perform word segmentation and word embedding (such as Word2Vec, BERT, etc.) to extract text information contained in the process (such as process name, process description, processing opinions, notes, etc.).

[0062] Step S306: Perform word segmentation and word embedding processing on the text information in the normalized data based on the BERT model to obtain text features.

[0063] In some embodiments, the normalized text information is input into the BERT model, and the tokenizer inside the BERT model will segment the text into words or subwords (tokens). For example, for the sentence "The procurement process needs to go through multiple layers of approval", it will be divided into tokens such as "procurement", "process", "need", "pass", "multi-layer" and "approval" based on BERT. Then, through the word embedding layer of the BERT model, each token is mapped to a high-dimensional vector. These vectors contain rich semantic information. These vector sequences constitute text features, which can capture the lexical meaning, grammatical structure and contextual semantic relationship in the process description.

[0064] Step S308: Statistically process the attribute information of the information to obtain statistical features.

[0065] Extract statistical information such as processing time, processing times, error rate, return rate, etc. of each processor in each link.

[0066] Step S310: extracting the association and influence characteristics between different process handlers in the organizational chart as network features based on the graph neural network.

[0067] In some embodiments, an enterprise's organizational chart is obtained, which contains information such as the roles, hierarchical relationships, and collaborative relationships of process handlers. The organizational chart is used as input to a graph neural network (GNN). GNN can learn the complex connections and influence characteristics between different process handlers by performing information propagation and aggregation operations on the graph structure.

[0068] Step S204: construct a model to be trained based on the pre-trained language model and the deep learning model.

[0069] In some embodiments, the pre-trained language model may be BERT, GPT, etc. The pre-trained language model has acquired rich semantic representation capabilities by pre-training on a large-scale corpus. These large models can be used for fine-tuning to adapt them to recommendation tasks in the process processing field.

[0070] In some embodiments, deep learning models can be used to capture the dependencies and timing features between process links. Deep learning models include but are not limited to convolutional neural networks (CNN), recurrent neural networks (RNN) and their variants (such as LSTM, GRU) and Transformer, etc. These models perform well in processing sequence data (such as time series data, text data).

[0071] In some embodiments, the fine-tuned pre-trained language model can be used as a front-end model of a deep learning model, and the output of the pre-trained language model can be used as an input of the deep learning model.

[0072] In some embodiments, the fine-tuned pre-trained language model and deep learning model can also be used as base learners to build a recommendation model.

[0073] In some embodiments, a suitable deep learning model can also be selected based on the characteristics of statistical features and network features and the nature of the recommendation task. If there are complex nonlinear relationships between text features, statistical features and network features, a multilayer perceptron (MLP) or a more complex neural network architecture (such as an MLP with multiple hidden layers) can be selected as a deep learning model to capture these relationships. If the temporal or sequential nature of the data is taken into account, a recurrent neural network (RNN) and its variants (such as LSTM, GRU) can also be selected as a deep learning model.

[0074] Step S206, training the to-be-trained model based on the text features, statistical features and network features to obtain a recommendation model, so as to recommend a processor for the to-be-processed process by the recommendation model.

[0075] In some embodiments, the extracted text features, statistical features, and network features are combined into training samples, and each training sample has a corresponding historical process handler label. For example, a training sample contains a text feature vector of a specific process (output by the BERT model), a statistical feature vector (a vector composed of features such as processing time and resource consumption), a network feature vector (a handler association and influence feature vector output by a graph neural network), and the identification of the person actually responsible for processing the process.

[0076] In some embodiments, the model training task for recommending process handlers can generally be regarded as a classification task (if the handler is a discrete category) or a regression task (if the selection of the handler is related to certain continuous values, such as processing efficiency scores, etc.). If it is a classification task, the cross entropy loss function can be selected; if it is a regression task, the mean square error loss function can be selected, etc.

[0077] In some embodiments, the optimizer may select Adam, Adagrad or other suitable optimizers.

[0078] In some embodiments, the training samples are input into the model to be trained, the prediction result of the model (i.e., the recommended process handler) is calculated through forward propagation, and then the error between the prediction result and the true label is calculated according to the loss function, and the optimizer is used for back propagation to update the parameters of the model (including the parameters of the pre-trained language model and the deep learning model).

[0079] In this embodiment, by parsing text features, statistical features and network features from historical process data, model training of the model to be trained is performed with comprehensive multi-dimensional information. The model to be trained is constructed based on a pre-trained language model and a deep learning model. The pre-trained language model and the deep learning model are constructed by introducing an artificial intelligence algorithm. The pre-trained language model can capture rich semantic information in the process description, and the deep learning model can capture the dependencies and timing characteristics between process links, thereby realizing the mining and precise analysis of historical process data to discover the association between process links and handlers. A recommendation model for process handlers is obtained based on model training, which can accurately identify personnel suitable for handling specific processes to ensure the reliability of the process handler recommendation and prevent the process from being assigned to handlers who may cause process bottlenecks, thereby improving the efficiency of process processing.

[0080] In one embodiment of the present disclosure, a model to be trained is constructed based on a pre-trained language model and a deep learning model, including:

[0081] The historical process data marked with recommended handlers are input into the pre-trained language model, and the difference between the recommendation results and the recommended handlers is calculated; the model parameters of the pre-trained language model are adjusted based on the difference, that is, the pre-trained language model is fine-tuned based on the historical process data to obtain a fine-tuned language model suitable for process management; the model to be trained is constructed based on the fine-tuned language model and the deep learning model.

[0082] In some embodiments, historical process data is input into a pre-trained language model. Since the pre-trained language model has been trained on large-scale general text data, it has a certain language understanding and representation ability. Under this premise, the model will generate corresponding prediction results based on the input process data, that is, recommended handlers. The difference between the recommended results predicted by the model and the actually labeled recommended handlers is calculated through appropriate evaluation indicators (such as cross-entropy loss function, if the recommended handler is a classification problem represented in probabilistic form). The difference reflects the inaccuracy of the model's current prediction. According to the calculated difference, an optimization algorithm (such as Adam optimizer) is used to adjust the parameters of the pre-trained language model. The goal of the adjustment is to gradually reduce the difference between the model's predicted results and the actual annotations. During the adjustment process, the model will learn specific patterns and features related to the recommended handlers in the historical process data to obtain a fine-tuned language model.

[0083] In this embodiment, by fine-tuning on a large amount of historical process data, the model learns the processing modes and handler recommendation rules for various types of processes. When combined with the deep learning model to construct a model to be trained, relevant information between other processes can be further integrated. The model can consider more comprehensive factors when recommending handlers to improve the reliability of the recommendation.

[0084] In one embodiment of the present disclosure, a model to be trained is constructed based on a fine-tuned language model and a deep learning model, including: using the fine-tuned language model as a front-end model and using the deep learning model as a back-end model to construct the model to be trained.

[0085] In some embodiments, the text feature vector output by the front-end fine-tuning language model is used as an input vector to the back-end deep learning model, so that the deep learning model outputs a predicted recommended handler.

[0086] In this embodiment, the fine-tuned language model has a strong ability to process text information. As a front-end model, it can deeply understand the text description related to the process and extract the semantic information therein, and the back-end deep learning model can effectively process other types of process features, such as statistical features and network features. It can capture the complex relationship between these features, so that the model training can comprehensively consider more information, thereby ensuring the reliability of the training effect.

[0087] like Figure 4 As shown, in one embodiment of the present disclosure, a model to be trained is trained based on text features, statistical features and network features to obtain a recommendation model, including:

[0088] Step S402: Select the corresponding loss function and optimizer according to the task type of the recommendation handler.

[0089] Among them, the task types include classification tasks and regression tasks. If it is a classification task, the cross entropy loss function can be used. If it is a regression task, the mean square error loss function can be selected.

[0090] Optimizers include but are not limited to Adam and AdagradRMSProp.

[0091] Step S404: input text features, statistical features and network features into a fine-tuned language model, and output a fused feature vector.

[0092] Among them, text features (extracted from text information such as process descriptions), statistical features (quantitative data such as process processing time and resource consumption) and network features (association information between processors) are input into the fine-tuned language model. The fine-tuned language model has been trained on data in a specific field and can process these different types of features to obtain fused features.

[0093] Step S406: input the fused feature vector into the deep learning model to capture the timing characteristics and dependencies between different historical processes based on the deep learning model, and output the recommended label.

[0094] Among them, the fused feature vector is input into the deep learning model, which captures the timing characteristics and dependencies between different historical processes according to its architecture (such as multi-layer perceptron, convolutional neural network, recurrent neural network, etc.).

[0095] Step S408, calculating the loss between the recommended label and the actual label of the recommender based on the loss function.

[0096] In some embodiments, based on a pre-selected loss function, the loss between the recommended label output by the deep learning model and the actual recommended handler label is calculated. The loss value reflects the degree of difference between the model prediction result and the actual situation. The selected optimizer is used to update the parameters of the deep learning model according to the loss value.

[0097] Step S410: Optimize the model parameters of the deep learning model based on the optimizer and the overfitting prevention strategy to obtain an optimized model.

[0098] Among them, overfitting prevention strategies include L1 / L2 regularization, Dropout, early stopping, etc. L1 / L2 regularization prevents the model from being too complex by adding a penalty term for the size of model parameters in the loss function; Dropout randomly discards some neurons during the training process to reduce the co-adaptation relationship between neurons; Early stopping stops training when the performance of the model on the validation set no longer improves. These methods are used to optimize model parameters and obtain an optimized model.

[0099] Step S412, verifying the performance of the optimization model based on the validation set to adjust the hyperparameters of the optimization model to obtain a recommended model.

[0100] Among them, an independent validation set data is used to evaluate the optimized model. By calculating performance indicators such as accuracy, recall, F1 value (for classification tasks) or mean square error (for regression tasks) on the validation set, the performance of the model on unseen data can be understood. The hyperparameters of the optimized model are adjusted according to the performance feedback of the validation set to obtain the recommended model.

[0101] In this embodiment, the loss function and optimizer are selected according to the task type of the recommendation handler to ensure the accuracy of the model training direction. Multiple features are input into the fine-tuned language model to output a fused feature vector, which fully integrates the key information in the text, statistics and network features. The deep learning model uses the fused vector to capture the temporal characteristics and dependencies of the historical process, thereby outputting more reasonable recommendation labels. The combination of the loss function, optimizer and verification scheme is conducive to improving the accuracy and efficiency of the obtained recommendation model.

[0102] In one embodiment of the present disclosure, a model to be trained is constructed based on a fine-tuned language model and a deep learning model, including: constructing the model to be trained by using the fine-tuned language model and the deep learning model as base learners in a tree model, respectively.

[0103] In some embodiments, a decision tree, a random forest, a gradient boosted tree (GBDT), etc. can be selected as the tree model.

[0104] In some embodiments, taking random forest as an example, the output feature vectors of the fine-tuned language model and the deep learning model are used as new features. For each training sample, its text features are input into the fine-tuned language model, and the statistical features and network features are input into the deep learning model to obtain the output vectors of the two models. These two vectors are then concatenated into a new feature vector. This new feature vector will serve as the input feature of each decision tree in the random forest. During the construction process of each decision tree, the best split features and split points are selected based on these new features to minimize the loss function (such as the Gini index or information gain in the classification task, the mean square error in the regression task, etc.), thereby constructing the entire random forest model as the model to be trained.

[0105] In this embodiment, a tree model (such as a random forest) is used to integrate two base learners. By combining the results of multiple decision trees, the variance of the model is reduced, the risk of overfitting is reduced, and the trained recommendation model can stably output recommendation results.

[0106] like Figure 5 As shown, in one embodiment of the present disclosure, a model to be trained is trained based on text features, statistical features and network features to obtain a recommendation model, including:

[0107] Step S502: Input text features, statistical features, and network features into a fine-tuned language model and a deep learning model to obtain first output features and second output features, respectively.

[0108] Among them, after the fine-tuned language model receives text, statistical and network features, it can use its internal neural network architecture to analyze the relationship between these features from a text perspective. For example, it can identify the connection between a specific process text description and certain statistical features (such as the association between urgent processes and short processing times) and network features (such as the extensive connection between processes involving multi-department collaboration and handlers). By learning and encoding these relationships, it outputs a first output feature vector containing rich semantic and contextual information.

[0109] The deep learning model processes the three input features according to its own structure (such as multi-layer perceptron, convolutional neural network or recurrent neural network, etc.) to further explore the mutual relationship. The second output feature vector can reflect the comprehensive understanding of the input features from the perspective of deep learning.

[0110] Step S504: concatenate the first output feature and the second output feature to obtain a concatenated feature.

[0111] Step S506: the tree model includes a decision tree, and the splitting features and splitting points of the decision tree are selected based on the splicing features.

[0112] In the process of building a decision tree, for each internal node, a split feature and a split point need to be selected to determine how to divide the data of the current node into two or more sub-nodes. Using the concatenated feature vector, the algorithm traverses each dimension of the vector and calculates the effect of using each dimension as a split feature on the data division.

[0113] Step S508, cross-validating the tree model of the generated decision tree, adjusting the parameters of the tree model based on the validation results, and obtaining a recommendation model.

[0114] Among them, cross-validation is used to evaluate the tree model of the generated decision tree, and the parameters of the tree model are adjusted according to the performance indicators (such as accuracy, recall rate, F1 value, mean square error, etc.) obtained on the validation set during the cross-validation process to obtain the recommended model.

[0115] In this embodiment, by combining text features, statistical features and network features, and using a fine-tuned language model and a deep learning model to process these features separately, and then fusing the results, the model can make multi-faceted considerations for recommendations to process handlers. Combined with the decision tree's ability to handle complex nonlinear relationships, the trained model can better understand more complex processing flows, thereby improving the reliability of the recommendation results.

[0116] like Figure 6 As shown, the method for recommending a process handler according to an embodiment of the present disclosure includes:

[0117] Step S602: In response to the generated process to be processed, extract process information in the process to be processed.

[0118] In some embodiments, when a new pending process is generated, the system will start an information extraction mechanism. This mechanism will deeply analyze various aspects of the pending process and obtain relevant process information from it. The process information covers a wide range, including the text description of the process (such as whether the process is about new product development, customer order processing, etc.), the type of process (whether it is a regular process or an emergency process), the business field involved in the process (such as finance, manufacturing, etc.), the requirements of each link of the process (such as the conditions of the approval link, the standards of the execution link, etc.), and some quantitative information related to the process (such as the estimated processing time, the number of resource requirements, etc.). By extracting this information, input data is provided for the subsequent recommendation model.

[0119] Step S604: input the process information into a recommendation model for process handlers to output recommended handlers for the process to be processed and corresponding recommendation probabilities, wherein the recommendation model is constructed based on a pre-trained language model and a deep learning model.

[0120] In some embodiments, after the extracted process information is input into the recommendation model, the text content in the process information is processed based on the trained pre-trained language model to extract the semantic features and grammatical structure information therein, and a comprehensive analysis is performed based on the trained deep learning model combined with these text features and other quantitative features to output a list of recommended handlers for the process and the recommendation probability corresponding to each handler.

[0121] Step S606: Acquire the status information of the recommendation handler, adjust the recommendation handler and the corresponding recommendation probability based on the status information, and generate a recommendation target.

[0122] Among them, the status information of the processor includes the degree of busyness, whether he is on duty and the authority level.

[0123] In some embodiments, during the recommendation process, multiple factors in actual business scenarios such as the status of the person handling the matter (such as busyness, holiday arrangements), authority level, etc. are taken into consideration, and the recommendation results are dynamically adjusted, thereby enhancing the flexibility and practicality of the recommendation system, ensuring that the recommendation results are more in line with the actual operational needs of the enterprise, and reducing recommendation errors caused by changes in personnel status.

[0124] In this embodiment, through automated data processing, model training and recommended processes, human intervention is greatly reduced and processing efficiency is improved, so that the process management system can effectively use the recommendation model in combination with process information to generate a preliminary list of recommended handlers and probabilities when a new process to be processed appears, and further optimize and adjust the recommendation results based on the real-time status information of the handler, so as to accurately determine the most suitable handler, improve the rationality and efficiency of process allocation, reduce process delays or improper processing caused by unreasonable allocation, and enhance the reliability and flexibility of the entire process management system to better adapt to complex and changing business environments.

[0125] In one embodiment of the present disclosure, in response to a generated process to be processed, process information in the process to be processed is extracted, including: in response to a generated process to be processed, extracting at least one of process description information in the process to be processed, processing information on the process, and association information between historical processors of the process to be processed as process information.

[0126] In one embodiment of the present disclosure, a recommendation model is generated based on a fine-tuned language model and a deep learning model connected in series, the fine-tuned language model is generated based on fine-tuning a pre-trained language model, and the process information is input into the recommendation model of the process handler to output a recommended handler for handling the process to be processed and a corresponding recommendation probability, including:

[0127] The process information is input into the fine-tuned language model, and a fused feature vector describing the process to be processed is output. The fused feature vector integrates text features, statistical features, and network features. The fused feature vector is input into the deep learning model to capture the dependency relationship between the process to be processed and the completed process based on the deep learning model, and the recommended handler of the process to be processed and the corresponding recommendation probability are output.

[0128] In one embodiment of the present disclosure, a recommendation model is generated based on respectively using a fine-tuned language model and a deep learning model as base learners in a tree model, the fine-tuned language model is generated based on fine-tuning a pre-trained language model, and the process information is input into the recommendation model of the process handler to output a recommended handler for processing the process to be processed and a corresponding recommendation probability, including:

[0129] The process information is input into the fine-tuning language model and the deep learning model to obtain the first output feature and the second output feature respectively; the first output feature and the second output feature are concatenated to obtain the concatenated feature; the concatenated feature is input into the decision tree in the tree model to output the recommended handler and the corresponding recommendation probability of the process to be processed.

[0130] In one embodiment of the present disclosure, it also includes: generating a display interface for recommending a target; in response to a selection operation on the interface, if a recommended target is selected, pushing the process to be processed to the terminal of the recommended target; if the recommended target is not selected, collecting the processor information specified by the user, and feeding back the specified processor information to the recommendation model for model optimization.

[0131] In some embodiments, a visual display interface is created based on the recommendation target (including a list of recommendation handlers and corresponding recommendation probabilities), and the recommendation information can be displayed in a list format, with each recommendation handler occupying a line, displaying their name, position, recommendation probability, and some brief personal information to facilitate user viewing and operation.

[0132] In some embodiments, if the user does not select a recommendation target, the system will provide an entry or method for the user to manually specify a handler. The user can enter the handler's name, work number and other information. After collecting the handler information specified by the user, it will be used as feedback data to optimize the recommendation model. That is, the new data will be incorporated into the model's training data to adjust the model's parameters so that the model can better meet the user's expectations and actual business conditions in the subsequent recommendation process. The collection of user feedback promotes the iterative optimization of the model, enabling the recommendation system to continue to learn and improve.

[0133] like Figure 7 As shown, a method for recommending a process handler according to another embodiment of the present disclosure includes:

[0134] Step S702, collect and organize historical process data of the process management system, and perform pre-processing.

[0135] Among them, preprocessing includes cleaning: removing duplicate data and erroneous data (such as incorrect names of processors, inconsistent time formats, etc.).

[0136] Preprocessing also includes deduplication: ensuring that data from different links of the same process are not counted repeatedly.

[0137] Preprocessing also includes formatting: unifying the format of time, date and other fields to ensure data consistency and comparability.

[0138] Step S704, extracting features from the preprocessed data to obtain text features, statistical features and network features.

[0139] Among them, the BERT model is used to embed words into texts such as process descriptions and processing opinions, and generate high-dimensional vector representations as text features.

[0140] Statistical information such as processing time, processing times, error rate, and return rate of each processor in each link is extracted as statistical features.

[0141] Using graph neural network (GNN) technology, the association and influence characteristics between managers are extracted as network features based on the internal organizational structure of the enterprise.

[0142] Step S706: construct a model to be trained based on the deep learning model and the fine-tuned language model in combination with an integrated learning framework.

[0143] Among them, the BERT model was selected as the basis, and the company's historical process data was used for fine-tuning to adapt it to the recommendation tasks of procurement process handlers. The LSTM model was combined to capture the timing characteristics and dependencies between process links. The random forest model was used to integrate the prediction results of BERT and LSTM to improve the accuracy and robustness of the recommendations.

[0144] Step S708, using the cross entropy loss function as the optimization target, and selecting the Adam optimizer for model training.

[0145] Among them, the model performance is evaluated through cross-validation, and hyperparameters such as learning rate and dropout rate are adjusted to prevent overfitting.

[0146] In addition, the model is retrained regularly to adapt to changes in corporate processes and personnel adjustments.

[0147] Step S710, detect the process to be processed, extract relevant information of the process, including the process type and link description, call the trained recommendation model for prediction, generate a list of recommended handlers and their recommendation probabilities, and display the recommendation results to the user in a list form.

[0148] Step S712, collecting feedback results from the processor on whether the user accepts or rejects the recommendation, using the user feedback as new training data, and retraining and optimizing the model.

[0149] In this embodiment, the intelligent process handler recommendation scheme based on historical process data can be put into practical application in process management such as enterprise procurement, which significantly improves the efficiency and accuracy of process management.

[0150] It should be noted that the above figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present disclosure, and are not intended to be limiting. It is easy to understand that the processes shown in the above figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0151] Refer to the following Figure 8 hereinafter, a recommendation model training device 800 according to an embodiment of the present disclosure is described. Figure 8The recommendation model training device 800 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0152] The recommendation model training device 800 is in the form of a hardware module or a software module. The components of the recommendation model training device 800 may include but are not limited to: a feature extraction module 802, which is used to extract features from the historical process data of the process management system to obtain text features, statistical features and network features, wherein the text features are used to characterize the description information of the process, the statistical features are used to characterize the processing information of the process, and the network features are used to characterize the association information between the historical processors of the process; a model construction module 804, which is used to construct a model to be trained based on a pre-trained language model and a deep learning model; a model training module 806, which is used to train the model to be trained based on text features, statistical features and network features to obtain a recommendation model, so that the recommendation model can recommend processors for the process to be processed.

[0153] Refer to the following Fig. 9 The process processing person recommendation device 900 according to the embodiment of the present disclosure is described. Fig. 9 The process handler recommendation device 900 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiment of the present disclosure.

[0154] The process handler recommendation device 900 is expressed in the form of a hardware module or a software module. The components of the process handler recommendation device 900 may include but are not limited to: an information extraction module 902, which is used to extract process information in the process to be processed in response to the generated process to be processed; a model processing module 904, which is used to input the process information into the process handler recommendation model to output the recommended handler and the corresponding recommendation probability for processing the process to be processed, wherein the recommendation model is constructed based on the pre-trained language model and the deep learning model; an adjustment module 906, which is used to obtain the status information of the recommended handler, to adjust the recommended handler and the corresponding recommendation probability based on the status information, and generate a recommendation target.

[0155] like Fig.10 As shown, the electronic device 1000 is in the form of a general computing device. The components of the electronic device 1000 may include but are not limited to: at least one processing unit 1010, at least one storage unit 1020, and a bus 1030 connecting different system components (including the storage unit 1020 and the processing unit 1010).

[0156] The storage unit stores program codes, which can be executed by the processing unit 1010, so that the processing unit 1010 performs the steps according to various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification. For example, the processing unit 1010 can perform the following steps: Figure 2 The scheme described in .

[0157] The storage unit 1020 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 10201 and / or a cache storage unit 10202 , and may further include a read-only storage unit (ROM) 10203 .

[0158] The storage unit 1020 may also include a program / utility 10204 having a set (at least one) of program modules 10205, such program modules 10205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0159] Bus 1030 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0160] The electronic device 1000 may also communicate with one or more external devices 7100 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 1000, and / or communicate with any device that enables the electronic device 1000 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 1050. In addition, the electronic device 1000 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 1060. As shown, the network adapter 1060 communicates with other modules of the electronic device 1000 via a bus 1030. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 1000, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0161] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or an electronic device, etc.) to execute the method according to the implementation of the present disclosure.

[0162] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the above method of the present specification is stored. In some possible implementations, various aspects of the present disclosure may also be implemented in the form of a program product, which includes a program code, and when the program product is run on an electronic device, the program code is used to enable the electronic device to perform the steps according to various exemplary implementations of the present disclosure described in the above "Exemplary Method" section of the present specification.

[0163] According to the program product for implementing the above method according to the embodiment of the present disclosure, it can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on an electronic device, such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, an apparatus or a device.

[0164] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0165] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0166] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0167] The program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the IoT terminal, as an independent software package, partially on the user computing device and partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0168] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0169] In addition, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0170] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or an electronic device, etc.) to execute the method according to the implementation of the present disclosure.

[0171] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the appended claims.

Claims

1. A recommendation model training method, characterized in that: include: Extracting features from the historical process data of the process management system to obtain text features, statistical features and network features, wherein the text features are used to characterize the description information of the process, the statistical features are used to characterize the processing information of the process, and the network features are used to characterize the association information between the historical processors of the process; Build the model to be trained based on the pre-trained language model and deep learning model; The model to be trained is trained based on the text features, the statistical features and the network features to obtain the recommendation model, so that the recommendation model recommends a processor for the process to be processed.

2. The recommendation model training method according to claim 1, characterized in that: Build the model to be trained based on the pre-trained language model and deep learning model, including: Inputting the historical process data marked with the recommended handler into the pre-trained language model, and calculating the difference between the recommendation result and the recommended handler; Adjusting model parameters of the pre-trained language model based on the difference to obtain a fine-tuned language model suitable for process management; The model to be trained is constructed based on the fine-tuning language model and the deep learning model.

3. The recommendation model training method according to claim 2, characterized in that: Building the model to be trained based on the fine-tuning language model and the deep learning model includes: The fine-tuning language model is used as the front-end model, and the deep learning model is used as the back-end model to construct the model to be trained.

4. The recommendation model training method according to claim 3, characterized in that: Performing model training on the model to be trained based on the text features, the statistical features, and the network features to obtain the recommendation model includes: Select the corresponding loss function and optimizer according to the task type of the recommendation handler; Inputting the text features, the statistical features and the network features into the fine-tuning language model, and outputting a fused feature vector; Inputting the fused feature vector into the deep learning model to capture the temporal features and dependencies between different historical processes based on the deep learning model, and outputting a recommended label; Calculating the loss between the recommended label and the real recommendation handler's label based on the loss function; and Optimizing the model parameters of the deep learning model based on the optimizer and the overfitting prevention strategy to obtain an optimized model; The performance of the optimization model is verified based on the validation set to adjust the hyperparameters of the optimization model to obtain the recommendation model.

5. The recommendation model training method according to claim 2, characterized in that: Constructing the model to be trained based on the fine-tuning language model and the deep learning model, including: The fine-tuning language model and the deep learning model are respectively used as base learners in a tree model to construct the model to be trained.

6. The recommendation model training method according to claim 5, characterized in that: Performing model training on the model to be trained based on the text features, the statistical features, and the network features to obtain the recommendation model includes: Inputting the text feature, the statistical feature and the network feature into the fine-tuning language model and the deep learning model to obtain a first output feature and a second output feature respectively; Concatenating the first output feature and the second output feature to obtain a concatenated feature; The tree model includes a decision tree, and a split feature and a split point of the decision tree are selected based on the splicing feature; The tree model used to generate the decision tree is cross-validated to adjust the parameters of the tree model based on the validation result to obtain the recommendation model.

7. The recommendation model training method according to claim 1, characterized in that: Feature extraction is performed on the historical process data of the process management system to obtain text features, statistical features and network features, including: Collecting the historical process data from the process management system, the historical process data including process type, process processor, processing information and supplementary information; Preprocessing the historical process data to obtain normalized data; Performing word segmentation and word embedding processing on the text information in the normalized data based on the BERT model to obtain the text features; Counting the attribute information of the processed information to obtain the statistical features; Based on the graph neural network, the association and influence characteristics between different process handlers in the organizational chart are extracted as the network features.

8. A method for recommending process handlers, characterized in that: include: In response to the generated process to be processed, extracting process information in the process to be processed; Inputting the process information into a recommendation model for process handlers to output recommended handlers for handling the process to be processed and corresponding recommendation probabilities, wherein the recommendation model is constructed based on a pre-trained language model and a deep learning model; The status information of the recommendation handler is obtained, so as to adjust the recommendation handler and the corresponding recommendation probability based on the status information and generate a recommendation target.

9. The method for recommending a process handler according to claim 8, characterized in that: In response to the generated to-be-processed process, extracting process information in the to-be-processed process includes: In response to the generated process to be processed, at least one of process description information in the process to be processed, process processing information, and association information between historical processors of the process to be processed is extracted as the process information.

10. The method for recommending a process handler according to claim 8, characterized in that: The recommendation model is generated based on a fine-tuning language model and a deep learning model connected in series, the fine-tuning language model is generated based on fine-tuning the pre-trained language model, the process information is input into the recommendation model of the process handler, so as to output the recommended handler for handling the process to be processed and the corresponding recommendation probability, including: Inputting the process information into the fine-tuning language model, and outputting a fused feature vector describing the process to be processed, wherein the fused feature vector fuses text features, statistical features, and network features; The fused feature vector is input into the deep learning model to capture the dependency relationship between the process to be processed and the completed process based on the deep learning model, and the recommended person to handle the process to be processed and the corresponding recommendation probability are output.

11. The method for recommending a process handler according to claim 8, characterized in that: The recommendation model is generated based on respectively using the fine-tuning language model and the deep learning model as base learners in the tree model, the fine-tuning language model is generated based on fine-tuning the pre-trained language model, and the process information is input into the recommendation model of the process handler to output the recommended handler for handling the process to be processed and the corresponding recommendation probability, including: Inputting the process information into the fine-tuning language model and the deep learning model to obtain a first output feature and a second output feature respectively; Concatenating the first output feature and the second output feature to obtain a concatenated feature; The splicing features are input into the decision tree in the tree model, and the recommended person to handle the process to be processed and the corresponding recommendation probability are output.

12. The method for recommending a process handler according to claim 8, characterized in that: Also includes: Generating a display interface of the recommended target; In response to a selection operation on the interface, if the recommended target is selected, the to-be-processed process is pushed to a terminal of the recommended target; If the recommendation target is not selected, the processor information specified by the user is collected, and the specified processor information is fed back to the recommendation model for model optimization.

13. A recommendation model training device, characterized in that: include: A feature extraction module is used to extract features from the historical process data of the process management system to obtain text features, statistical features and network features. The text features are used to characterize the description information of the process, the statistical features are used to characterize the processing information of the process, and the network features are used to characterize the association information between the historical processors of the process; Model building module, used to build the model to be trained based on the pre-trained language model and deep learning model; The model training module is used to perform model training on the model to be trained based on the text features, the statistical features and the network features to obtain the recommendation model, so that the recommendation model can recommend a processor for the process to be processed.

14. A device for recommending process handlers, characterized in that: include: An information extraction module, configured to extract process information in the process to be processed in response to the generated process to be processed; A model processing module, used for inputting the process information into a recommendation model for process handlers to output a recommended handler for processing the process to be processed and a corresponding recommendation probability, wherein the recommendation model is constructed based on a pre-trained language model and a deep learning model; The adjustment module is used to obtain the status information of the recommendation handler, adjust the recommendation handler and the corresponding recommendation probability based on the status information, and generate a recommendation target.

15. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to execute the recommendation model training method described in any one of claims 1 to 7 and / or the process handler recommendation method described in any one of claims 8 to 12 by executing the executable instructions.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the recommendation model training method described in any one of claims 1 to 7 and / or the process handler recommendation method described in any one of claims 8 to 12.

17. A computer program product having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the recommendation model training method described in any one of claims 1 to 7 and / or the process handler recommendation method described in any one of claims 8 to 12.

Images