Work order dispatching method and device, processor and electronic device

Abstract

The present application discloses a method and device for dispatching work orders, a processor and an electronic device, and relates to the field of artificial intelligence. The method includes: obtaining a first potential feature of multiple target words in a target work order, wherein the target work order is a work order to be dispatched, and the target words are words other than stop words in the target work order; combining the fully connected layer and the normalized exponential function of the neural network, based on the first potential feature, a plurality of probability values ​​are calculated, wherein the probability value is the probability value of the target work order corresponding to each organizational level; according to the plurality of probability values ​​and the first preset threshold, the target organizational level corresponding to the target work order is determined; based on the target organizational level, the target organization corresponding to the target work order is determined, and the target work order is dispatched to the target organization. Through the present application, the problem of low accuracy in dispatching work orders in the related art is solved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and more specifically, to a method and device for dispatching work orders, a processor, and an electronic device. Background Art

[0002] In the related art, the processing mode of work orders is generally through manual means, flowing downward step by step according to the organizational hierarchy, which undoubtedly increases the work order processing time. In addition, the manual assignment of work orders not only requires strong professional qualities, but also consumes a lot of energy. Moreover, the work order handler needs to pay attention to the work order assignment status regularly every day to reduce the work order response time. Therefore, processing the work orders to be assigned brings an additional burden to the relevant personnel. Therefore, it is an inevitable demand to invent a solution for intelligently assigning work orders. However, most of the solutions for intelligently assigning work orders in the related art lack the capture of deep semantic information of the work order text, which leads to low accuracy in assigning work orders.

[0003] Currently, no effective solution has been proposed to address the problem of low accuracy in dispatching work orders in related technologies. Summary of the Invention

[0004] The main purpose of this application is to provide a method and device for dispatching work orders, a processor and an electronic device to solve the problem of low accuracy in dispatching work orders in related technologies.

[0005] To achieve the above-mentioned objectives, according to one aspect of the present application, a method for assigning work orders is provided. The method comprises: obtaining first latent features of multiple target words in a target work order, wherein the target work order is the work order to be assigned, and the target words are words in the target work order excluding stop words; combining a fully connected layer of a neural network and a normalized exponential function, and calculating multiple probability values ​​based on the first latent features, wherein the probability values ​​are probability values ​​of the target work order corresponding to each organizational level; determining the target organizational level corresponding to the target work order based on the multiple probability values ​​and a first preset threshold; determining the target organizational level corresponding to the target work order based on the target organizational level, and assigning the target work order to the target organizational level.

[0006] Furthermore, based on the multiple probability values ​​and the first preset threshold, determining the target organizational level corresponding to the target work order includes: determining the two largest probability values ​​among the multiple probability values; subtracting the two largest probability values ​​among the multiple probability values ​​to obtain a target numerical value; judging whether the target numerical value is greater than the first preset threshold; if the target numerical value is greater than the first preset threshold, determining the largest probability value among the multiple probability values, and taking the organizational level corresponding to the largest probability value as the target organizational level; if the target numerical value is not greater than the first preset threshold, obtaining the second potential features of the multiple target words in the target work order; combining the fully connected layer of the neural network and the normalized exponential function, and based on the second potential features, determining the target organizational level corresponding to the target work order.

[0007] Furthermore, obtaining the first potential features of multiple target words in the target work order includes: filtering the target characters in the target work order to obtain a first work order, wherein the target characters are at least one of the following: special characters and useless characters; filtering the stop words in the first work order based on a stop word list to obtain a second work order; performing word segmentation processing on each word in the second work order based on a first word dictionary to obtain multiple word vectors; processing each word vector to obtain a word embedding matrix; inputting the word embedding matrix into the ELMO model for processing to obtain the first potential features of multiple target words in the target work order.

[0008] Furthermore, obtaining the second potential features of multiple target words in the target work order includes: obtaining a weight matrix for each institutional level based on the appearance of each target word in the target work order in each second word dictionary and the TF-IDF matrix of each institutional level, wherein the second word dictionary is the word dictionary corresponding to the historical work orders of each institutional level; performing dot multiplication operations on the weight matrix of each institutional level and the word embedding matrix respectively to obtain a weight matrix for each institutional level; constructing an image matrix based on the weight matrix of each institutional level and the first potential features; performing convolution processing on the image matrix, and obtaining the second potential features of multiple target words in the target work order based on the maximum pooling method.

[0009] Furthermore, the word embedding matrix is ​​input into the ELMO model for processing to obtain the first potential features of multiple target words in the target work order, including: using the forward LSTM layer and the backward LSTM layer in the ELMO model to process the target work order respectively to obtain a third potential feature and a fourth potential feature; and performing weighted summation on the word embedding matrix, the third potential feature, and the fourth potential feature to obtain the first potential features of multiple target words in the target work order.

[0010] Furthermore, before obtaining the weight matrix of each institutional level based on the occurrence of each target word in the target work order in each second word dictionary and the TF-IDF matrix of each institutional level, the method also includes: obtaining multiple historical work orders from which special characters, useless characters and stop words are filtered out; dividing each historical work order into corresponding institutional levels; performing word segmentation processing on the historical work orders of each institutional level and establishing multiple second word dictionaries; and using the TF-IDF formula to obtain the TF-IDF matrix of each institutional level based on each second word dictionary.

[0011] Furthermore, based on the target organization hierarchy, determining the target organization corresponding to the target work order includes: extracting key information containing the target organization hierarchy from the target work order; performing text similarity matching on the key information and the organization tree of the target organization hierarchy to obtain similarity; judging whether the similarity is greater than a second preset threshold; if the similarity is greater than the second preset threshold, determining the target organization corresponding to the target work order from the organization tree; if the similarity is not greater than the second preset threshold, taking the organization with the highest hierarchy as the target organization corresponding to the target work order.

[0012] To achieve the above-mentioned purpose, according to another aspect of the present application, a work order dispatching device is provided. The device comprises: a first acquisition unit, configured to acquire first potential features of multiple target words in a target work order, wherein the target work order is a work order to be dispatched, and the target words are words other than stop words in the target work order; a first calculation unit, configured to combine a fully connected layer of a neural network and a normalized exponential function to calculate multiple probability values ​​based on the first potential features, wherein the probability values ​​are probability values ​​of the target work order corresponding to each organization level; a first determination unit, configured to determine the target organization level corresponding to the target work order based on the multiple probability values ​​and a first preset threshold; and a first processing unit, configured to determine the target organization corresponding to the target work order based on the target organization level, and dispatch the target work order to the target organization.

[0013] Furthermore, the first determination unit includes: a first determination module, used to determine the two largest probability values ​​among the multiple probability values; a first calculation module, used to subtract the two largest probability values ​​among the multiple probability values ​​to obtain a target value; a first judgment module, used to judge whether the target value is greater than the first preset threshold; a second determination module, used to determine the largest probability value among the multiple probability values ​​if the target value is greater than the first preset threshold, and use the organizational level corresponding to the largest probability value as the target organizational level; a first acquisition module, used to acquire the second potential features of multiple target words in the target work order if the target value is not greater than the first preset threshold; a third determination module, used to combine the fully connected layer of the neural network and the normalized exponential function, and determine the target organizational level corresponding to the target work order based on the second potential features.

[0014] Furthermore, the first acquisition unit includes: a first filtering module, used to filter the target characters in the target work order to obtain a first work order, wherein the target characters are at least one of the following: special characters and useless characters; a second filtering module, used to filter the stop words in the first work order based on a stop word list to obtain a second work order; a first processing module, used to perform word segmentation processing on each word in the second work order based on a first word dictionary to obtain multiple word vectors; a second processing module, used to process each word vector to obtain a word embedding matrix; a third processing module, used to input the word embedding matrix into the ELMO model for processing to obtain the first potential features of multiple target words in the target work order.

[0015] Furthermore, the first acquisition module includes: a first determination submodule, which is used to obtain the weight matrix of each institutional level based on the appearance of each target word in the target work order in each second word dictionary and the TF-IDF matrix of each institutional level, wherein the second word dictionary is the word dictionary corresponding to the historical work orders of each institutional level; a first operation submodule, which is used to perform dot multiplication operation on the weight matrix of each institutional level and the word embedding matrix respectively to obtain the weight matrix of each institutional level; a first construction submodule, which is used to construct an image matrix based on the weight matrix of each institutional level and the first potential feature; a second determination submodule, which is used to perform convolution processing on the image matrix and obtain the second potential features of multiple target words in the target work order based on the maximum pooling device.

[0016] Furthermore, the third processing module includes: a first processing sub-module, used to process the target work order using the forward LSTM layer and the backward LSTM layer in the ELMO model respectively to obtain a third potential feature and a fourth potential feature; a third determination sub-module, used to perform weighted summation on the word embedding matrix, the third potential feature and the fourth potential feature to obtain the first potential feature of multiple target words in the target work order.

[0017] Furthermore, the device also includes: a second acquisition unit, which is used to obtain multiple historical work orders that filter out special characters, useless characters and stop words before obtaining the weight matrix of each organizational level based on the appearance of each target word in the target work order in each second word dictionary and the TF-IDF matrix of each organizational level; a first division unit, which is used to divide each historical work order into the corresponding organizational level; a first establishment unit, which is used to perform word segmentation processing on the historical work orders of each organizational level and establish multiple second word dictionaries; a second determination unit, which is used to adopt the TF-IDF formula to obtain the TF-IDF matrix of each organizational level based on each second word dictionary.

[0018] Furthermore, the first processing unit includes: a first extraction module, used to extract key information of the target organization level contained in the target work order; a fourth determination module, used to perform text similarity matching between the key information and the organization tree of the target organization level to obtain similarity; a second judgment module, used to judge whether the similarity is greater than a second preset threshold; a fifth determination module, used to determine the target organization corresponding to the target work order from the organization tree if the similarity is greater than the second preset threshold; and a sixth determination module, used to use the organization with the highest level as the target organization corresponding to the target work order if the similarity is not greater than the second preset threshold.

[0019] In order to achieve the above-mentioned object, according to another aspect of the present application, a processor is provided, wherein the processor is used to run a program, wherein the program executes any one of the above-mentioned work order dispatching methods when running.

[0020] In order to achieve the above-mentioned purpose, according to another aspect of the present application, an electronic device is provided, which includes one or more processors and a memory, and the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement any one of the above-mentioned work order dispatching methods.

[0021] Through the present application, the following steps are adopted: obtaining the first potential features of multiple target words in the target work order, wherein the target work order is the work order to be assigned, and the target words are the words in the target work order excluding stop words; combining the fully connected layer and the normalized exponential function of the neural network, based on the first potential features, multiple probability values ​​are calculated, wherein the probability values ​​are the probability values ​​of the target work order corresponding to each organizational level; determining the target organizational level corresponding to the target work order based on the multiple probability values ​​and the first preset threshold; determining the target organizational level corresponding to the target work order based on the target organizational level, and assigning the target work order to the target organization, thereby solving the problem of low accuracy in assigning work orders in related technologies. By combining the fully connected layer and normalized exponential function of the neural network, based on the potential features of multiple words in the obtained work orders to be processed, the probability value of the target work order corresponding to each organizational level is calculated. Based on the calculated probability value and the preset threshold, the organizational level corresponding to the work order to be processed is determined. Then, based on the organizational level corresponding to the work order to be processed, the organizational level corresponding to the work order to be processed is determined, and the work order to be processed is assigned to the organizational level, thereby achieving the effect of improving the accuracy of work order assignment. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0023] Figure 1 This is a flowchart of a method for dispatching work orders according to an embodiment of the present application;

[0024] Figure 2 This is a flowchart of the text preprocessing module training in the embodiment of the present application;

[0025] Figure 3 This is a flowchart of the prediction stage in the embodiment of the present application;

[0026] Figure 4 This is a flowchart of the word embedding module in an embodiment of the present application;

[0027] Figure 5 This is a flowchart of a work order potential dependency learning module in an embodiment of the present application;

[0028] Figure 6 is a flowchart of the allocation level prediction module in an embodiment of the present application;

[0029] Figure 7 This is a flowchart of the decision-making module in an embodiment of the present application;

[0030] Figure 8 This is a flowchart of the reinforcement learning module in an embodiment of the present application;

[0031] Figure 9 This is a flow chart of the mechanism matching module in an embodiment of the present application;

[0032] Figure 10 This is a flowchart of the automated work order dispatching module in an embodiment of the present application;

[0033] Figure 11 is a flowchart of an optional work order dispatching method provided according to an embodiment of the present application;

[0034] Figure 12 is a schematic diagram of a work order dispatching device provided according to an embodiment of the present application;

[0035] Figure 13 is a schematic diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0036] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0037] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0038] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present application described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0039] It should be noted that the relevant information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display and analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties. For example, an interface is set up between this system and the relevant user or organization. Before obtaining relevant information, it is necessary to send an acquisition request to the aforementioned user or organization through the interface, and obtain the relevant information after receiving the consent information fed back by the aforementioned user or organization.

[0040] For ease of description, some nouns or terms involved in the embodiments of the present application are explained below:

[0041] The ELMO model is a bidirectional language model, a new type of deeply contextualized word representation. Its full name stands for Embedding from Language Model. It uses bidirectional LSTMs as its basic network component, modeling complex word features (such as syntax and semantics) and the changes in word context (i.e., modeling polysemy). Our word vectors are a function of the internal state of a deep bidirectional language model (BiLM), pre-trained on a large text corpus. When talking about word vectors, we inevitably think of word2vec, as its concept of word vectors has significantly advanced the development of natural language processing (NLP). The ELMO model's key approach is to first train a complete language model, then use this language model to process the text to be trained to generate the corresponding word vectors. Therefore, the ELMO model can generate different word vectors for the same word in different sentences.

[0042] Long Short-Term Memory (LSTM) is a time-recurrent neural network designed to solve the long-term dependency problem of general recurrent neural networks (RNNs). All RNNs have a chain-like form of repeated neural network modules.

[0043] The normalized exponential function, or Softmax function, can "compress" a K-dimensional vector z containing any real number into another K-dimensional real vector σ(z), so that each element ranges between (0, 1) and the sum of all elements is 1. This function is often used in multi-classification problems.

[0044] In a fully connected layer, each node is connected to all nodes in the previous layer, integrating the previously extracted features. Due to its fully connected nature, fully connected layers generally have the most parameters.

[0045] TF-IDF (term frequency–inverse document frequency) is a common weighting technique used in information retrieval and data mining. TF stands for term frequency, and IDF stands for inverse document frequency. In other words, TF-IDF is a statistical method used to assess the importance of a word to a collection of documents or a document in a corpus. The importance of a word increases proportionally with the number of times it appears in a document, but decreases inversely with the frequency of its occurrence in the corpus. Various forms of TF-IDF weighting are often used by search engines as a measure or rating of the relevance between documents and user queries.

[0046] Word2vec is a related model used to generate word vectors, which is used for training to reconstruct linguistic word texts.

[0047] TextCNN is an algorithm that uses convolutional neural networks to classify text.

[0048] CNN stands for Convolutional Neural Networks, and its Chinese name is Convolutional Neural Network. It recognizes text in images through convolution.

[0049] RPA technology, or Robotic Process Automation, is a digital technology that automates the work of programs and systems by mimicking user operations on computers.

[0050] The present invention will be described below in conjunction with preferred implementation steps. Figure 1 This is a flow chart of a method for dispatching work orders according to an embodiment of the present application. Figure 1 As shown, the method includes the following steps:

[0051] Step S101 : obtaining first potential features of a plurality of target words in a target work order, wherein the target work order is a work order to be assigned, and the target words are words other than stop words in the target work order.

[0052] For example, the first latent feature may be the latent grammatical and semantic features of the words in the work order. Stop words in the work order to be assigned are first filtered out, and then the latent grammatical and semantic information of the remaining words in the work order is obtained.

[0053] In step S102 , a plurality of probability values ​​are calculated based on the first latent feature by combining the fully connected layer of the neural network and the normalized exponential function, wherein the probability value is the probability value of the target work order corresponding to each organization level.

[0054] For example, by obtaining the latent grammatical and semantic information of the remaining words in the work order, excluding stop words, and then performing a fully connected layer and softmax normalization, the probability of each processing level corresponding to the work order is output. Furthermore, the target work order can be a work order to be assigned at a financial institution, and the organizational level can be branches, sub-branches, and outlets.

[0055] Step S103: determining the target organization level corresponding to the target work order based on the multiple probability values ​​and the first preset threshold.

[0056] For example, based on the probability value of each processing level corresponding to the output work order and the preset threshold, the level to which the work order to be assigned belongs is determined, that is, whether the work order to be assigned is assigned to a branch, a sub-branch, or an outlet.

[0057] Step S104: Determine the target organization corresponding to the target work order based on the target organization hierarchy, and assign the target work order to the target organization.

[0058] For example, after determining that a work order is to be assigned to a branch, the specific branch to which the work order is to be assigned is determined. The assignment of branches and outlets is similar to that of branches. That is, after determining that a work order is to be assigned to a branch, the specific branch to which the work order is to be assigned is determined; after determining that a work order is to be assigned to an outlet, the specific outlet to which the work order is to be assigned is determined.

[0059] Through the above steps S101 to S104, by combining the fully connected layer and the normalized exponential function of the neural network, based on the potential features of multiple words in the obtained work orders to be processed, the probability value of the target work order corresponding to each organizational level is calculated, and based on the calculated probability value and the preset threshold, the organizational level corresponding to the work order to be processed is determined, and then based on the organizational level corresponding to the work order to be processed, the organization corresponding to the work order to be processed is determined, and the work order to be processed is assigned to the organization, thereby achieving the effect of improving the accuracy of assigned work orders.

[0060] In order to quickly and accurately determine the TF-IDF matrix of each organizational level, in the work order dispatching method provided in the embodiment of the present application, the TF-IDF matrix of each organizational level can also be determined by the following steps: obtaining multiple historical work orders that filter out special characters, useless characters, and stop words; dividing each historical work order into corresponding organizational levels; performing word segmentation processing on the historical work orders of each organizational level and establishing multiple second-word dictionaries; using the TF-IDF formula, based on each second-word dictionary, obtaining the TF-IDF matrix of each organizational level.

[0061] For example, the automated work order dispatching solution may include a text preprocessing module, a word embedding module, a work order potential dependency capture module, a decision module, a reinforcement learning module, a dispatching level prediction module, an organization matching module, and an automated work order dispatching module. In addition, Figure 2 This is a flowchart of the text preprocessing module training in the embodiment of the present application. Figure 2 As shown in the figure, the text preprocessing module first filters out special and useless characters in work orders and removes stop words based on a stop word list. The Jieba word segmentation tool then segments historical work orders and creates a word list. Based on the actual processing level of historical work orders, the work orders are divided into branches, sub-branches, and outlets. The TF-IDF matrix is ​​calculated for each level using the TF-IDF method for subsequent reinforcement learning.

[0062] Through the above solution, a word list and a TF-IDF matrix within each level can be established based on historical work orders, which can be facilitated by subsequent word embedding modules and reinforcement learning modules.

[0063] In order to quickly and accurately obtain the first potential features of multiple target words in the target work order, in the work order dispatching method provided in the embodiment of the present application, the first potential features of multiple target words in the target work order can also be obtained by the following steps: filtering the target characters in the target work order to obtain a first work order, wherein the target characters are at least one of the following: special characters and useless characters; based on the stop word list, filtering the stop words in the first work order to obtain a second work order; based on the first word dictionary, performing word segmentation processing on each word in the second work order to obtain multiple word vectors; processing each word vector to obtain a word embedding matrix; inputting the word embedding matrix into the ELMO model for processing to obtain the first potential features of multiple target words in the target work order.

[0064] For example, Figure 3 This is a flow chart of the prediction stage in the embodiment of the present application. Figure 3 As shown, special and useless characters in the newly input work order are filtered, and stop words are removed based on the stop word list. Then, with the help of the jieba word segmentation tool, the newly input work order is segmented and word vectors are established. In addition, Figure 4This is a flowchart of the word embedding module in the embodiment of the present application. Figure 4 As shown in the figure, word vectors are segmented based on the established word list to obtain the segmented word vectors. Then, based on the word2vec model, these segmented word vectors are represented as a work order embedding matrix for subsequent model building and learning. Finally, the work order embedding matrix is ​​input into the trained ELMO model to obtain the latent grammatical and semantic features of the words in the new input work order.

[0065] Through the above scheme, the potential grammatical and semantic features of the words in the newly input work order can be quickly and accurately obtained through the work order embedding matrix and the ELMO model.

[0066] In order to quickly and accurately obtain the first potential features of multiple target words in the target work order, in the work order dispatching method provided in the embodiment of the present application, the first potential features of multiple target words in the target work order can also be obtained by the following steps: the target work order is processed using the forward LSTM layer and the backward LSTM layer in the ELMO model respectively to obtain the third potential feature and the fourth potential feature; the word embedding matrix, the third potential feature and the fourth potential feature are weighted and summed to obtain the first potential features of multiple target words in the target work order.

[0067] For example, Figure 2 This is a flowchart of the text preprocessing module training in the embodiment of the present application. Figure 2 As shown, before modeling the latent dependencies of work order text, the work order latent dependency capture module first loads an ELMO model that has been pre-trained on a large Chinese corpus. It then retrains the pre-trained ELMO model for a specified number of rounds using outbound call conversations and historical work order data, fine-tuning the relevant parameters of the pre-trained model. Subsequently, the pre-processed work order text is embedded and input into the fine-tuned ELMO model to generate contextual semantic representations. This model models the latent dependencies of the text and helps the dispatch level prediction module learn the similarities between work orders at the same dispatch level.

[0068] In addition, ELMO is based on the bidirectional language model (BiLM) and is constructed using the bidirectional long short-term memory network (BiLSTM) structure. Its training goal is to maximize the probability of the forward and backward language model predicting correctly. Specifically, given a sequence of length N (t1, ...t N ), the forward language model is based on the historical position sequence (t1, ..., t k-1 ), use forward LSTM modeling to learn and predict the next position as t k The probability of maximizing, while the backward language model is the opposite, based on the future position sequence (t k+1 ,...,t N ). Use backward LSTM modeling to learn and predict the historical position k as t kThe probability of maximizing, so the training goal of the ELMO model is to maximize the log-likelihood probability:

[0069]

[0070] Among them, Θ x is the embedding vector input to the ELMO model, is the potential vector representation of the output of the forward (backward) LSTM layer, Θ s It is the context matrix after softmax function normalization.

[0071] and, Figure 5 This is a flowchart of the work order potential dependency learning module in the embodiment of the present application, such as Figure 5 As shown in the figure, in the ELMO model, after word segmentation, work orders are processed using word2vec to generate embedding vectors. These embedding vectors are then fed into a two-layer BiLSTM structure to model syntactic and semantic features. The model outputs are the work order embedding vector, the hidden state output by the forward LSTM, and the weighted average of the hidden state output by the backward LSTM. Furthermore, to better represent the textual semantics of work orders, a corpus built from financial knowledge and historical work order data is pre-trained into the ELMO model. Based on experimental results, a small number of training rounds is determined. Based on the determined number of rounds, ELMO is retrained, relevant parameters are fine-tuned, and the trained model is saved. When a new work order request is input into this module, the fine-tuned model is loaded and outputs the latent semantic expression features of the work order text.

[0072] In summary, by training the ELMO model, the pre-trained ELMO model is used to obtain the potential grammatical and semantic features of the words in the new input work order.

[0073] In order to quickly and accurately determine the target organizational level corresponding to the target work order, in the work order dispatching method provided in the embodiment of the present application, the target organizational level corresponding to the target work order can also be determined by the following steps: determining the two largest probability values ​​among multiple probability values; subtracting the two largest probability values ​​among the multiple probability values ​​to obtain a target numerical value; judging whether the target numerical value is greater than a first preset threshold; if the target numerical value is greater than the first preset threshold, determining the largest probability value among the multiple probability values, and taking the organizational level corresponding to the largest probability value as the target organizational level; if the target numerical value is not greater than the first preset threshold, obtaining the second potential features of multiple target words in the target work order; combining the fully connected layer and the normalized exponential function of the neural network, based on the second potential features, determining the target organizational level corresponding to the target work order.

[0074] For example, Figure 6 This is a flowchart of the allocation level prediction module in the embodiment of the present application. Figure 6As shown, the dispatch level prediction module consists of a fully connected layer and a softmax function. It is responsible for mapping the captured latent dependency expressions into the dispatch level of the work order and generating the model's prediction. It is important to note that when the dispatch level prediction module receives input from the work order text latent dependency capture module, its output is not necessarily used as the model's final predicted dispatch level. Instead, the weight values ​​and predicted level after softmax processing are passed to the decision module. The decision module then determines whether to directly output the level (outputting the organization level with the highest probability value as the level corresponding to the work order to be dispatched) or enter the reinforcement learning module.

[0075] in addition, Figure 7 This is a flow chart of the decision module in the embodiment of the present application. Figure 7 As shown, the decision module evaluates the output of the dispatch level prediction module. If the result is credible, the dispatch level predicted by the dispatch level prediction module is directly output, and the organization level with the highest probability value is used as the level corresponding to the work order to be dispatched. Otherwise, the reinforcement learning module is used for further classification. Specifically, the credibility function f(s1, s2) is defined as:

[0076]

[0077] Where s1 and s2 are the two values ​​with the highest probability after the softmax function calculation in the dispatch level prediction module, and σ is the threshold. When |s1-s2| ≥ σ, the output of f(s1, s2) is 1, indicating that the model's prediction is clear and the model's dispatch level can be directly classified and output. If the output is 0, indicating that the model's prediction is less reliable, the reinforcement learning module is used to further explore the relationship between the keywords in the work order and the classification level.

[0078] Through the above scheme, the organizational level corresponding to the work order to be assigned can be determined quickly and accurately based on the probability value.

[0079] In order to quickly and accurately obtain the second potential features of multiple target words in the target work order, in the work order dispatching method provided in the embodiment of the present application, the second potential features of multiple target words in the target work order can also be obtained through the following steps: based on the appearance of each target word in the target work order in each second word dictionary and the TF-IDF matrix of each organizational level, a weight matrix for each organizational level is obtained, wherein the second word dictionary is the word dictionary corresponding to the historical work orders of each organizational level; the weight matrix of each organizational level is dot-multiplied with the word embedding matrix to obtain a weighted matrix for each organizational level; an image matrix is ​​constructed based on the weight matrix of each organizational level and the first potential features; the image matrix is ​​convolved, and based on the maximum pooling method, the second potential features of multiple target words in the target work order are obtained.

[0080] For example, Figure 8 This is a flow chart of the enhanced learning module in the embodiment of the present application. Figure 8 As shown in the figure, in the reinforcement learning module, the TF-IDF method is used to calculate the influence of terms within a ticket on the dispatch level, and a different weight matrix is ​​established for each dispatch level. When the main model's discriminability for the input ticket is insufficient, we use the weight matrix for each level to weight the learned ticket embeddings. This further enhances the importance of "highly discriminative" terms to the model's classification, helping the model better distinguish between the dispatch levels of tickets.

[0081] Specifically, the filtered historical work orders are grouped based on the work order assignment level, all words are extracted from the corresponding group i (i∈1, 2, 3), and the TF-IDF matrix of the corresponding category is established. The elements in the matrix describe the importance of the word in category i and are calculated by the following formula.

[0082]

[0083]

[0084]

[0085] For any group i, and |d i | Represents the word t i Occurrence count and total number of words, |O i | indicates work order o i the number of Indicates that the group contains the word t i The number of work orders, and |T i |Indicates the number of all words in the work order.

[0086] Based on TF-IDF matrix The weight matrix of the input work order can be generated by the following rules. When a word in the input work order appears in a work order of the corresponding category (dispatching level) i, we assign it the corresponding category TF-IDF matrix M i The maximum value of the column corresponding to the word position in the work order. For a word in the work order that does not appear in the work order of category i but appears in the work order of other category k (k≠i), we assign it M i The minimum of all column maximums in . If a word does not appear in the existing TF-IDF matrix M, it is assigned the mean of the identified words in the weight matrix and is considered a "non-influential" word. In this way, a corresponding weight matrix is ​​established for each input ticket, and this is dot-producted with the embedding vector output by the word embedding module to obtain the weighted ticket embedding vector for each word in different categories.

[0087] Afterwards, the output of the ELMO layer is stacked with the multi-category embedding weighted matrix established above and input into the multi-channel textCNN to capture the influence of words on the different importance of different categories. After dimensionality reduction through maximum pooling, the learned potential dependency expression is output and passed into the dispatch level prediction module to output the predicted dispatch level.

[0088] In summary, through the reinforcement learning module, the potential features of the words in the modeled work orders to be assigned can be output quickly and accurately.

[0089] In order to quickly and accurately determine the target organization corresponding to the target work order, in the work order dispatching method provided in the embodiment of the present application, the target organization corresponding to the target work order can also be determined by the following steps: extracting the key information of the target organization level contained in the target work order; performing text similarity matching on the key information and the organization tree of the target organization level to obtain the similarity; judging whether the similarity is greater than a second preset threshold; if the similarity is greater than the second preset threshold, determining the target organization corresponding to the target work order from the organization tree; if the similarity is not greater than the second preset threshold, taking the organization with the highest level as the target organization corresponding to the target work order.

[0090] For example, the text similarity calculation module passes the processed work order content and organization tree into this module, processes the work order content based on regular expressions, extracts key information, and performs similarity matching calculations with the organization tree. The similarity matching uses the SequenceMatcher method (used to compare two characters and return data based on their similarity). If the calculated similarity is greater than the specified threshold, the corresponding organization name is output.

[0091] in addition, Figure 9 This is a flow chart of the mechanism matching module in the embodiment of the present application. Figure 9 As shown, the organization matching module uses different methods to match work order content with work order processing organizations based on the differences in the dispatch levels output by the model. The specific rules are as follows:

[0092] (1) The organizational level of this module is input as branch, and the work orders are screened according to the established expert rules and the business category to which the work orders belong. If the screening is successful, the corresponding department name is output according to the screening category.

[0093] (2) When the input level is the branch, the module will first extract the key information including the branch name in the work order based on regular expressions and other means, and calculate the match with the branch organization tree under the branch through the text similarity calculation module. If the matching branch name is output, the branch department to which the work order belongs will be further filtered based on expert rules, the business category of the work order, etc. If the screening is successful, the corresponding branch department name will be output.

[0094] (3) The input level is the branch. This module uses the same processing method to extract key information containing the branch and calculates the text similarity with the branch organization tree. If a match is successful, the corresponding branch name is output.

[0095] (4) If this module fails to successfully match the organization, the default output is the name of the department that handles the branch-level work order.

[0096] also, Figure 10 This is a flow chart of the automated work order dispatching module in the embodiment of the present application. Figure 10 As shown, the automated work order dispatching module is used by RPA to receive the output content from the organization matching module, locate the position of the dispatching operation element, and automatically dispatch the work order to the corresponding organization level for processing.

[0097] The above solution allows work orders to be quickly and accurately assigned to the appropriate processing organization based on the organization level to which they are assigned. Furthermore, RPA technology can replace manual work order dispatching, reducing the time staff spend processing work orders and alleviating the workload on the business.

[0098] For example, Figure 11 This is a flow chart of an optional work order dispatching method provided in accordance with an embodiment of the present application, such as Figure 11 As shown in the figure, word2Vec is used to embed the work order content, and multi-layer BiLSTM is used to capture the deep latent semantic and grammatical features of the embedded work order, so as to solve the impact of polysemy on the model's work order assignment. A decision-making mechanism is introduced. Based on the model's judgment on the credibility of the work order assignment level prediction, a weighted embedding matrix is ​​established for each assignment level. This is superimposed with the latent grammatical and semantic expression of the multi-layer BiLSTM structure and passed into textCNN for learning and modeling. Finally, through the fully connected layer and softmax normalization processing, the processing level corresponding to the work order is output. With the help of RPA technology, manual processing of work orders is simulated to realize intelligent dispatch of work orders, thereby saving business personnel time and reducing the business burden.

[0099] In summary, the work order dispatching method provided in the embodiment of the present application obtains the first potential features of multiple target words in the target work order, wherein the target work order is the work order to be dispatched, and the target words are the words in the target work order excluding stop words; combining the fully connected layer and the normalized exponential function of the neural network, based on the first potential features, multiple probability values ​​are calculated, wherein the probability value is the probability value of the target work order corresponding to each organizational level; based on the multiple probability values ​​and the first preset threshold, the target organizational level corresponding to the target work order is determined; based on the target organizational level, the target organization corresponding to the target work order is determined, and the target work order is dispatched to the target organization, thereby solving the problem of low accuracy in dispatching work orders in related technologies. By combining the fully connected layer and normalized exponential function of the neural network, based on the potential features of multiple words in the obtained work orders to be processed, the probability value of the target work order corresponding to each organizational level is calculated. Based on the calculated probability value and the preset threshold, the organizational level corresponding to the work order to be processed is determined. Then, based on the organizational level corresponding to the work order to be processed, the organizational level corresponding to the work order to be processed is determined, and the work order to be processed is assigned to the organizational level, thereby achieving the effect of improving the accuracy of work order assignment.

[0100] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0101] The present application also provides a work order dispatching device. It should be noted that the work order dispatching device of the present application can be used to execute the work order dispatching method provided in the present application. The work order dispatching device provided in the present application is introduced below.

[0102] Figure 12 Schematic diagram of a work order dispatching device according to an embodiment of the present application. Figure 12 As shown, the device includes: a first acquiring unit 1201 , a first calculating unit 1202 , a first determining unit 1203 and a first processing unit 1204 .

[0103] Specifically, the first acquisition unit 1201 is configured to acquire first potential features of a plurality of target words in a target work order, wherein the target work order is a work order to be assigned, and the target words are words other than stop words in the target work order;

[0104] A first calculation unit 1202 is configured to calculate a plurality of probability values ​​based on the first latent feature by combining a fully connected layer of a neural network and a normalized exponential function, wherein the probability value is a probability value of the target work order corresponding to each organization level;

[0105] A first determining unit 1203 is configured to determine a target organization level corresponding to the target work order based on multiple probability values ​​and a first preset threshold;

[0106] The first processing unit 1204 is configured to determine a target organization corresponding to the target work order based on the target organization hierarchy, and assign the target work order to the target organization.

[0107] In summary, the work order dispatching device provided in the embodiment of the present application obtains the first potential features of multiple target words in the target work order through the first acquisition unit 1201, wherein the target work order is the work order to be dispatched, and the target words are the words other than stop words in the target work order; the first calculation unit 1202 combines the fully connected layer and the normalized exponential function of the neural network, and calculates multiple probability values ​​based on the first potential features, wherein the probability value is the probability value of the target work order corresponding to each organizational level; the first determination unit 1203 determines the target organizational level corresponding to the target work order based on the multiple probability values ​​and the first preset threshold; the first processing unit 1204 determines the target organization corresponding to the target work order based on the target organizational level, and dispatches the target work order to the target organization, thereby solving the problem of low accuracy in dispatching work orders in related technologies. By combining the fully connected layer and normalized exponential function of the neural network, based on the potential features of multiple words in the obtained work orders to be processed, the probability value of the target work order corresponding to each organizational level is calculated. Based on the calculated probability value and the preset threshold, the organizational level corresponding to the work order to be processed is determined. Then, based on the organizational level corresponding to the work order to be processed, the organizational level corresponding to the work order to be processed is determined, and the work order to be processed is assigned to the organizational level, thereby achieving the effect of improving the accuracy of work order assignment.

[0108] Optionally, in the work order dispatching device provided in the embodiment of the present application, the first determination unit includes: a first determination module, used to determine the two largest probability values ​​among multiple probability values; a first calculation module, used to subtract the two largest probability values ​​among the multiple probability values ​​to obtain a target value; a first judgment module, used to judge whether the target value is greater than a first preset threshold; a second determination module, used to determine the largest probability value among the multiple probability values ​​if the target value is greater than the first preset threshold, and take the organizational level corresponding to the largest probability value as the target organizational level; a first acquisition module, used to acquire the second potential features of multiple target words in the target work order if the target value is not greater than the first preset threshold; a third determination module, used to combine the fully connected layer of the neural network and the normalized exponential function to determine the target organizational level corresponding to the target work order based on the second potential features.

[0109] Optionally, in the work order dispatching device provided in an embodiment of the present application, the first acquisition unit includes: a first filtering module, used to filter target characters in the target work order to obtain a first work order, wherein the target characters are at least one of the following: special characters and useless characters; a second filtering module, used to filter stop words in the first work order based on a stop word list to obtain a second work order; a first processing module, used to perform word segmentation processing on each word in the second work order based on a first word dictionary to obtain multiple word vectors; a second processing module, used to process each word vector to obtain a word embedding matrix; and a third processing module, used to input the word embedding matrix into the ELMO model for processing to obtain first potential features of multiple target words in the target work order.

[0110] Optionally, in the work order dispatching device provided in the embodiment of the present application, the first acquisition module includes: a first determination submodule, which is used to obtain the weight matrix of each organizational level based on the appearance of each target word in the target work order in each second word dictionary and the TF-IDF matrix of each organizational level, wherein the second word dictionary is the word dictionary corresponding to the historical work orders of each organizational level; a first operation submodule, which is used to perform dot multiplication operation on the weight matrix of each organizational level and the word embedding matrix respectively to obtain the weight matrix of each organizational level; a first construction submodule, which is used to construct an image matrix based on the weight matrix of each organizational level and the first potential feature; a second determination submodule, which is used to perform convolution processing on the image matrix and obtain the second potential features of multiple target words in the target work order based on the maximum pooling device.

[0111] Optionally, in the work order dispatching device provided in the embodiment of the present application, the third processing module includes: a first processing sub-module, used to process the target work order using the forward LSTM layer and the backward LSTM layer in the ELMO model respectively to obtain the third latent feature and the fourth latent feature; a third determination sub-module, used to perform weighted summation on the word embedding matrix, the third latent feature and the fourth latent feature to obtain the first latent feature of multiple target words in the target work order.

[0112] Optionally, in the work order dispatching device provided in the embodiment of the present application, the device also includes: a second acquisition unit, used to obtain multiple historical work orders that filter out special characters, useless characters and stop words before obtaining the weight matrix of each organizational level based on the appearance of each target word in the target work order in each second word dictionary and the TF-IDF matrix of each organizational level; a first division unit, used to divide each historical work order into the corresponding organizational level; a first establishment unit, used to perform word segmentation processing on the historical work orders of each organizational level and establish multiple second word dictionaries; a second determination unit, used to adopt the TF-IDF formula to obtain the TF-IDF matrix of each organizational level based on each second word dictionary.

[0113] Optionally, in the work order dispatching device provided in the embodiment of the present application, the first processing unit includes: a first extraction module, used to extract key information of the target organization level contained in the target work order; a fourth determination module, used to match the key information with the organization tree of the target organization level for text similarity to obtain similarity; a second judgment module, used to judge whether the similarity is greater than a second preset threshold; a fifth determination module, used to determine the target organization corresponding to the target work order from the organization tree if the similarity is greater than the second preset threshold; and a sixth determination module, used to use the organization with the highest level as the target organization corresponding to the target work order if the similarity is not greater than the second preset threshold.

[0114] The work order dispatching device includes a processor and a memory. The above-mentioned first acquisition unit 1201, first calculation unit 1202, first determination unit 1203 and first processing unit 1204 are all stored in the memory as program units, and the processor executes the above-mentioned program units stored in the memory to realize corresponding functions.

[0115] The processor contains a kernel, which retrieves the corresponding program unit from memory. You can configure one or more kernels, and adjust kernel parameters to improve the accuracy of work order dispatch.

[0116] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0117] An embodiment of the present invention provides a processor, which is used to run a program, wherein the work order dispatching method is executed when the program is run.

[0118] like Figure 13 As shown, an embodiment of the present invention provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, the following steps are implemented: obtaining a first potential feature of multiple target words in a target work order, wherein the target work order is a work order to be assigned, and the target words are words other than stop words in the target work order; combining the fully connected layer and the normalized exponential function of the neural network, based on the first potential feature, calculating multiple probability values, wherein the probability values ​​are probability values ​​of the target work order corresponding to each organization level; determining the target organization level corresponding to the target work order based on the multiple probability values ​​and a first preset threshold; determining the target organization corresponding to the target work order based on the target organization level, and assigning the target work order to the target organization.

[0119] When the processor executes the program, the following steps are also implemented: based on the multiple probability values ​​and the first preset threshold, determining the target organization level corresponding to the target work order includes: determining the two largest probability values ​​among the multiple probability values; subtracting the two largest probability values ​​among the multiple probability values ​​to obtain a target value; judging whether the target value is greater than the first preset threshold; if the target value is greater than the first preset threshold, determining the largest probability value among the multiple probability values, and taking the organization level corresponding to the largest probability value as the target organization level; if the target value is not greater than the first preset threshold, obtaining the second potential features of multiple target words in the target work order; combining the fully connected layer of the neural network and the normalized exponential function, based on the second potential features, determining the target organization level corresponding to the target work order.

[0120] When the processor executes the program, the following steps are also implemented: obtaining the first potential features of multiple target words in the target work order includes: filtering the target characters in the target work order to obtain a first work order, wherein the target characters are at least one of the following: special characters and useless characters; based on a stop word list, filtering the stop words in the first work order to obtain a second work order; based on a first word dictionary, performing word segmentation processing on each word in the second work order to obtain multiple word vectors; processing each word vector to obtain a word embedding matrix; inputting the word embedding matrix into the ELMO model for processing to obtain the first potential features of multiple target words in the target work order.

[0121] When the processor executes the program, the following steps are also implemented: obtaining the second potential features of multiple target words in the target work order includes: obtaining the weight matrix of each institutional level based on the appearance of each target word in the target work order in each second word dictionary and the TF-IDF matrix of each institutional level, wherein the second word dictionary is the word dictionary corresponding to the historical work orders of each institutional level; performing dot multiplication operations on the weight matrix of each institutional level and the word embedding matrix respectively to obtain the weight matrix of each institutional level; constructing an image matrix based on the weight matrix of each institutional level and the first potential features; performing convolution processing on the image matrix, and obtaining the second potential features of multiple target words in the target work order based on the maximum pooling method.

[0122] When the processor executes the program, the following steps are also implemented: inputting the word embedding matrix into the ELMO model for processing to obtain the first potential features of multiple target words in the target work order, including: using the forward LSTM layer and the backward LSTM layer in the ELMO model to process the target work order respectively to obtain the third potential feature and the fourth potential feature; performing weighted summation on the word embedding matrix, the third potential feature and the fourth potential feature to obtain the first potential features of multiple target words in the target work order.

[0123] When the processor executes the program, the following steps are also implemented: before obtaining the weight matrix of each organizational level based on the appearance of each target word in the target work order in each second word dictionary and the TF-IDF matrix of each organizational level, the method also includes: obtaining multiple historical work orders from which special characters, useless characters and stop words are filtered out; dividing each historical work order into corresponding organizational levels; performing word segmentation processing on the historical work orders of each organizational level and establishing multiple second word dictionaries; and using the TF-IDF formula to obtain the TF-IDF matrix of each organizational level based on each second word dictionary.

[0124] When the processor executes the program, the following steps are also implemented: based on the target organization hierarchy, determining the target organization corresponding to the target work order includes: extracting key information containing the target organization hierarchy from the target work order; performing text similarity matching on the key information and the organization tree of the target organization hierarchy to obtain similarity; judging whether the similarity is greater than a second preset threshold; if the similarity is greater than the second preset threshold, determining the target organization corresponding to the target work order from the organization tree; if the similarity is not greater than the second preset threshold, taking the organization with the highest hierarchy as the target organization corresponding to the target work order.

[0125] The devices in this article can be servers, PCs, PADs, mobile phones, etc.

[0126] The present application also provides a computer program product, which, when executed on a data processing device, is suitable for executing a program initialized with the following method steps: obtaining a first potential feature of multiple target words in a target work order, wherein the target work order is a work order to be assigned, and the target words are words other than stop words in the target work order; combining a fully connected layer and a normalized exponential function of a neural network, based on the first potential feature, calculating multiple probability values, wherein the probability values ​​are probability values ​​of the target work order corresponding to each organizational level; determining the target organizational level corresponding to the target work order based on the multiple probability values ​​and a first preset threshold; determining the target organizational level corresponding to the target work order based on the target organizational level, and assigning the target work order to the target organizational level.

[0127] When executed on a data processing device, it is also suitable for executing a program initialized with the following method steps: determining the target organization level corresponding to the target work order based on the multiple probability values ​​and a first preset threshold value, including: determining the two largest probability values ​​among the multiple probability values; subtracting the two largest probability values ​​among the multiple probability values ​​to obtain a target value; judging whether the target value is greater than the first preset threshold value; if the target value is greater than the first preset threshold value, determining the largest probability value among the multiple probability values, and taking the organization level corresponding to the largest probability value as the target organization level; if the target value is not greater than the first preset threshold value, obtaining a second potential feature of multiple target words in the target work order; combining the fully connected layer of the neural network and the normalized exponential function, and determining the target organization level corresponding to the target work order based on the second potential feature.

[0128] When executed on a data processing device, it is also suitable for executing a program initialized with the following method steps: obtaining the first potential features of multiple target words in a target work order includes: filtering the target characters in the target work order to obtain a first work order, wherein the target characters are at least one of the following: special characters and useless characters; based on a stop word list, filtering the stop words in the first work order to obtain a second work order; based on a first word dictionary, performing word segmentation processing on each word in the second work order to obtain multiple word vectors; processing each word vector to obtain a word embedding matrix; inputting the word embedding matrix into the ELMO model for processing to obtain the first potential features of multiple target words in the target work order.

[0129] When executed on a data processing device, it is also suitable for executing a program initialized with the following method steps: obtaining the second potential features of multiple target words in the target work order includes: based on the appearance of each target word in the target work order in each second word dictionary and the TF-IDF matrix of each institutional level, obtaining the weight matrix of each institutional level, wherein the second word dictionary is the word dictionary corresponding to the historical work orders of each institutional level; performing dot multiplication operations on the weight matrix of each institutional level and the word embedding matrix respectively to obtain the weight matrix of each institutional level; constructing an image matrix based on the weight matrix of each institutional level and the first potential features; performing convolution processing on the image matrix, and obtaining the second potential features of multiple target words in the target work order based on the maximum pooling method.

[0130] When executed on a data processing device, it is also suitable for executing an initialized program having the following method steps: inputting the word embedding matrix into the ELMO model for processing to obtain the first potential features of multiple target words in the target work order, including: using the forward LSTM layer and the backward LSTM layer in the ELMO model to process the target work order respectively to obtain a third potential feature and a fourth potential feature; performing weighted summation on the word embedding matrix, the third potential feature and the fourth potential feature to obtain the first potential features of multiple target words in the target work order.

[0131] When executed on a data processing device, it is also suitable for executing an initialized program having the following method steps: before obtaining a weight matrix for each organizational level based on the occurrence of each target word in the target work order in each second word dictionary and the TF-IDF matrix of each organizational level, the method further includes: obtaining multiple historical work orders from which special characters, useless characters, and stop words are filtered out; dividing each historical work order into corresponding organizational levels; performing word segmentation processing on the historical work orders of each organizational level and establishing multiple second word dictionaries; and using the TF-IDF formula to obtain the TF-IDF matrix for each organizational level based on each second word dictionary.

[0132] When executed on a data processing device, it is also suitable for executing a program initialized with the following method steps: based on the target organization hierarchy, determining the target organization corresponding to the target work order includes: extracting key information of the target organization hierarchy contained in the target work order; performing text similarity matching on the key information and the organization tree of the target organization hierarchy to obtain similarity; judging whether the similarity is greater than a second preset threshold; if the similarity is greater than the second preset threshold, determining the target organization corresponding to the target work order from the organization tree; if the similarity is not greater than the second preset threshold, taking the organization with the highest hierarchy as the target organization corresponding to the target work order.

[0133] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0134] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0135] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0136] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0137] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0138] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0139] Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0140] It should also be noted that the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, commodity, or apparatus that includes a series of elements includes not only those elements but also other elements not explicitly listed, or includes elements inherent to such process, method, commodity, or apparatus. In the absence of further limitations, an element defined by the phrase "comprises a ..." does not exclude the presence of other identical elements in the process, method, commodity, or apparatus that includes the element.

[0141] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0142] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for dispatching a work order, characterized in that: include: Obtaining first potential features of a plurality of target words in a target work order, wherein the target work order is a work order to be assigned, and the target words are words other than stop words in the target work order; Combining a fully connected layer of a neural network and a normalized exponential function, based on the first latent feature, a plurality of probability values ​​are calculated, wherein the probability value is a probability value of the target work order corresponding to each organization level; Determining a target organization level corresponding to the target work order based on the multiple probability values ​​and a first preset threshold; Determine the target organization corresponding to the target work order based on the target organization hierarchy, and assign the target work order to the target organization; Determining a target organization level corresponding to the target work order based on the multiple probability values ​​and a first preset threshold includes: Determining the two largest probability values ​​among the multiple probability values; Subtract the two largest probability values ​​from the multiple probability values ​​to obtain a target value; Determining whether the target value is greater than the first preset threshold; If the target value is greater than the first preset threshold, determining the maximum probability value among the multiple probability values, and using the organization level corresponding to the maximum probability value as the target organization level; If the target value is not greater than the first preset threshold, obtaining second potential features of multiple target words in the target work order; Determining a target organization level corresponding to the target work order based on the second latent feature by combining the fully connected layer of the neural network and the normalized exponential function; The method further includes processing the target work order to obtain a word embedding matrix; Wherein, obtaining the second potential features of the plurality of target words in the target work order includes: Based on the occurrence of each target word in the target work order in each second word dictionary and the TF-IDF matrix of each organization level, a weight matrix for each organization level is obtained, wherein the second word dictionary is the word dictionary corresponding to the historical work orders of each organization level; Performing a dot multiplication operation on the weight matrix of each organizational level and the word embedding matrix to obtain a weight matrix for each organizational level; constructing an image matrix according to the weighted matrix of each organizational level and the first latent feature; Performing convolution processing on the image matrix and obtaining second latent features of multiple target words in the target work order based on a maximum pooling method; The method further includes locating and assigning the position of the operation elements through robotic process automation technology, and automatically assigning the work order to the corresponding organizational level for processing.

2. The method according to claim 1, characterized in that The first potential features of multiple target words in the target work order are obtained as follows: Filtering target characters in the target work order to obtain a first work order, wherein the target characters are at least one of the following: special characters and useless characters; Filtering the stop words in the first work order based on a stop word list to obtain a second work order; Based on the first word dictionary, each word in the second work order is segmented to obtain multiple word vectors; Process each word vector to obtain the word embedding matrix; The word embedding matrix is ​​input into the ELMO model for processing to obtain the first potential features of multiple target words in the target work order.

3. The method according to claim 2, characterized in that The word embedding matrix is ​​input into the ELMO model for processing, and the first potential features of the target words in the target work order are obtained, including: The target work order is processed using the forward LSTM layer and the backward LSTM layer in the ELMO model to obtain a third latent feature and a fourth latent feature; A weighted sum is performed on the word embedding matrix, the third latent feature, and the fourth latent feature to obtain first latent features of multiple target words in the target work order.

4. The method according to claim 1, wherein Before obtaining a weight matrix for each organization level based on the occurrence of each target word in the target work order in each second word dictionary and the TF-IDF matrix for each organization level, the method further includes: Get multiple historical tickets that filter out special characters, useless characters, and stop words; Divide each historical work order into the corresponding organizational level; Perform word segmentation on historical work orders at each organizational level and establish multiple second-word dictionaries; Using the TF-IDF formula, based on each second-word dictionary, the TF-IDF matrix of each institutional level is obtained.

5. The method according to claim 1, characterized in that Determining the target organization corresponding to the target work order based on the target organization hierarchy includes: Extracting key information of the target organization level contained in the target work order; Performing text similarity matching between the key information and the organization tree of the target organization level to obtain similarity; Determining whether the similarity is greater than a second preset threshold; If the similarity is greater than the second preset threshold, determining the target organization corresponding to the target work order from the organization tree; If the similarity is not greater than the second preset threshold, the organization with the highest level is used as the target organization corresponding to the target work order.

6. A work order dispatching device, characterized in that: include: a first acquiring unit, configured to acquire first potential features of a plurality of target words in a target work order, wherein the target work order is a work order to be assigned, and the target words are words other than stop words in the target work order; a first calculation unit, configured to calculate a plurality of probability values ​​based on the first latent feature by combining a fully connected layer of a neural network and a normalized exponential function, wherein the probability values ​​are probability values ​​of the target work order corresponding to each organization level; A first determining unit, configured to determine a target organization level corresponding to the target work order based on the multiple probability values ​​and a first preset threshold; A first processing unit is configured to determine a target organization corresponding to the target work order based on the target organization hierarchy, and assign the target work order to the target organization; The first determination unit includes: a first determination module for determining the two largest probability values ​​among a plurality of probability values; a first calculation module for subtracting the two largest probability values ​​among the plurality of probability values ​​to obtain a target value; a first judgment module for judging whether the target value is greater than a first preset threshold; a second determination module for determining the largest probability value among the plurality of probability values ​​if the target value is greater than the first preset threshold, and using the organization level corresponding to the largest probability value as the target organization level; a first acquisition module for acquiring second potential features of a plurality of target words in a target work order if the target value is not greater than the first preset threshold; and a third determination module for determining a target organization level corresponding to the target work order based on the second potential features in combination with a fully connected layer of a neural network and a normalized exponential function; The device is further configured to process the target work order to obtain a word embedding matrix; The first acquisition module includes: a first determination submodule, configured to obtain a weight matrix for each organizational level based on the occurrence of each target word in the target work order in each second word dictionary and the TF-IDF matrix of each organizational level, wherein the second word dictionary is the word dictionary corresponding to the historical work orders of each organizational level; a first operation submodule, configured to perform a dot multiplication operation on the weight matrix of each organizational level and the word embedding matrix respectively to obtain a weight matrix for each organizational level; a first construction submodule, configured to construct an image matrix based on the weight matrix of each organizational level and the first latent feature; a second determination submodule, configured to perform convolution processing on the image matrix and obtain the second latent features of multiple target words in the target work order based on a maximum pooling device; The device is also used to locate and assign operation elements through robotic process automation technology, and automatically assign work orders to the corresponding organizational level for processing.

7. A processor, characterized in that: The processor is configured to run a program, wherein the program, when running, executes the work order dispatching method according to any one of claims 1 to 5.

8. An electronic device, characterized in that: It includes one or more processors and a memory, wherein the memory is used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the work order dispatching method described in any one of claims 1 to 5.

Images