Method and System for Accurately Identifying Voice Customer Service Intent Based on Deep Learning

Abstract

The present invention relates to the technical field of intelligent voice customer service, and a method and system for accurately identifying the intention of a voice customer service based on deep learning, including: determining whether there is only one intersection tree node in the subsequence of intersection tree nodes. If not, identifying multiple groups of associated customer voice word sets containing the subsequence of intersection tree nodes, and constructing multiple groups of branched tree node sequences on the last intersection tree node. If so, constructing multiple groups of branched tree node sequences on the last intersection tree node according to multiple groups of iterative customer voice word sets to obtain a customer voice word tree, collecting the customer voice word trees corresponding to each voice customer service intention to obtain a customer voice word tree forest, and performing deep learning recognition of the voice customer service intention according to the set of overlapping tree nodes to obtain the current customer service intention. The present invention can improve the recognition accuracy and recognition speed of the current intelligent customer service in the aspect of voice customer service intention recognition.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent voice customer service, and in particular, to a method and system for accurately identifying the intent of a voice customer service based on deep learning. Background Art

[0002] With the rapid development of artificial intelligence technology, voice interaction has become one of the mainstream ways of human-computer interaction. In the field of intelligent customer service, a voice customer service system realizes automated question answering, business handling, and problem solving by identifying the intent in a user's voice command, significantly improving service efficiency and user experience.

[0003] Traditional voice customer service systems have problems such as low accuracy in intent recognition, weak ability to understand multi-round conversations, and sensitivity to noise interference in complex scenarios. Therefore, current intelligent customer services have problems of low recognition accuracy and slow recognition speed in voice customer service intent recognition. Summary of the Invention

[0004] The present invention provides a method and system for accurately identifying the intent of a voice customer service based on deep learning, and its main purpose is to improve the recognition accuracy and recognition speed of current intelligent customer services in voice customer service intent recognition.

[0005] To achieve the above object, a method for accurately identifying the intent of a voice customer service based on deep learning provided by the present invention includes:

[0006] Obtain a set of customer voice texts corresponding to the intent of a voice customer service, perform word segmentation processing on the set of customer voice texts to obtain multiple sets of initial customer voice word sets;

[0007] Screen multiple sets of initial customer voice word sets according to a preset initial voice repetition word order to obtain multiple sets of target customer voice word sets;

[0008] Construct an intersection tree node sequence according to the initial voice repetition word order;

[0009] Identify multiple sets of intersection tree node subsequences of the intersection tree node sequence;

[0010] Extract intersection tree node subsequences in multiple sets of intersection tree node subsequences in turn according to a preset node decreasing order;

[0011] Judge whether there is only one intersection tree node in the intersection tree node subsequence;

[0012] If there is not only one intersection tree node in the intersection tree node subsequence, then identify multiple sets of associated customer voice word sets including the intersection tree node subsequence in multiple sets of target customer voice word sets;

[0013] Identify the last intersection tree node in the intersection tree node subsequence, and construct multiple branched tree node sequences on the last intersection tree node according to multiple groups of associated customer voice word sets;

[0014] Remove multiple groups of associated customer voice word sets from multiple groups of target customer voice word sets to obtain multiple groups of iterative customer voice word sets;

[0015] Update multiple groups of target customer voice word sets with multiple groups of iterative customer voice word sets, and return the above steps of sequentially extracting intersection tree node subsequences in multiple groups of intersection tree node subsequences according to the preset node decreasing order;

[0016] If there is only one intersection tree node in the intersection tree node subsequence, construct multiple branched tree node sequences on the last intersection tree node according to multiple groups of iterative customer voice word sets to obtain a customer voice word tree;

[0017] Collect the customer voice word trees corresponding to each voice customer intention to obtain a customer voice word tree forest;

[0018] Identify the overlapping tree node sets between the pre-acquired current customer voice word set and each customer voice word tree in the customer voice word tree forest;

[0019] Perform deep learning recognition of the voice customer intention according to the overlapping tree node sets to obtain the current customer intention, where the deep learning recognition is based on a pre-trained target convolutional neural network.

[0020] Optionally, the step of screening multiple groups of initial customer voice word sets according to a preset initial voice repetition word order to obtain multiple groups of target customer voice word sets includes:

[0021] Sequentially extract initial customer voice word sets from the multiple groups of initial customer voice word sets;

[0022] Judge whether the initial customer voice word set contains one or more initial voice repetition words in the initial voice repetition word order;

[0023] If the initial customer voice word set does not contain one or more initial voice repetition words in the initial voice repetition word order, return the above step of sequentially extracting initial customer voice word sets from the multiple groups of initial customer voice word sets;

[0024] If the initial customer voice word set contains one or more initial voice repetition words in the initial voice repetition word order, use the initial customer voice word set as the target customer voice word set;

[0025] Judge whether the extraction of the initial customer voice word set is completed;

[0026] If the initial customer voice word sets are not completely extracted, return to the above steps of sequentially extracting the initial customer voice word sets from the multiple groups of initial customer voice word sets;

[0027] If the initial customer voice word sets are completely extracted, multiple groups of target customer voice word sets are obtained.

[0028] Optionally, before screening the multiple groups of initial customer voice word sets according to the preset initial voice repetition word order to obtain multiple groups of target customer voice word sets, the method further includes:

[0029] Obtain the customer voice union word set of the multiple groups of initial customer voice word sets, where the customer voice union word set refers to the union of each group of initial customer voice word sets;

[0030] Sequentially extract customer voice union words from the customer voice union word set;

[0031] Count the number of word set groups in the multiple groups of initial customer voice word sets that contain the customer voice union word;

[0032] Judge whether the number of word set groups is greater than 1;

[0033] If the number of word set groups is not greater than 1, return to the above step of sequentially extracting customer voice union words from the customer voice union word set;

[0034] If the number of word set groups is greater than 1, use the customer voice union word as the initial voice repetition word to obtain the initial voice repetition word set;

[0035] Sort the initial voice repetition word set according to the number of word set groups of each initial voice repetition word in the initial voice repetition word set to obtain the initial voice repetition word order, where the initial voice repetition word order is sorted in descending order of the number of word set groups.

[0036] Optionally, constructing the intersection tree node sequence according to the initial voice repetition word order includes:

[0037] Sequentially extract initial voice repetition words from the initial voice repetition word order;

[0038] On the preset root node, use the initial voice repetition word to construct an iterative intersection tree node, where the node distance between the iterative intersection tree node and the root node is the preset node spacing;

[0039] Judge whether the initial voice repetition word is the last initial voice repetition word in the initial voice repetition word order;

[0040] If the initial voice repetition word is not the last initial voice repetition word in the initial voice repetition word order, update the root node using the iterative intersection tree node, and return the above steps of sequentially extracting the initial voice repetition words in the initial voice repetition word order;

[0041] If the initial voice repetition word is the last initial voice repetition word in the initial voice repetition word order, obtain the intersection tree node sequence.

[0042] Optionally, the multiple groups of intersection tree node subsequences for identifying the intersection tree node sequence include:

[0043] Obtain the initial node number;

[0044] Determine the initial grouped tree node in the intersection tree node sequence according to the initial node number, where the initial node number is 2;

[0045] Identify the sequence of precursor tree nodes of the initial grouped tree node, where the sequence of precursor tree nodes refers to the tree node sequence composed of the intersection tree nodes in the intersection tree node sequence that are in front of the initial grouped tree node;

[0046] Take the sequence of precursor tree nodes as the intersection tree node subsequence;

[0047] Perform an incremental update on the initial node number, where the incremental update means adding one to the initial node number;

[0048] Obtain the total number of intersection tree nodes in the intersection tree node sequence, calculate the target tree node number according to the total number of intersection tree nodes, where the difference between the target tree node number and the total number of intersection tree nodes is 1 and the target tree node number is greater than the total number of intersection tree nodes;

[0049] Judge whether the iterative node number is equal to the target tree node number;

[0050] If the iterative node number is not equal to the target tree node number, return the above step of determining the initial grouped tree node in the intersection tree node sequence according to the initial node number;

[0051] If the iterative node number is equal to the target tree node number, gather all the intersection tree node subsequences to obtain multiple groups of intersection tree node subsequences.

[0052] Optionally, the constructing of multiple groups of branched tree node sequences on the last intersection tree node according to multiple groups of associated customer voice word sets includes:

[0053] Identify the target customer voice word order corresponding to the intersection tree node subsequence;

[0054] Remove the target customer voice word order from the associated customer voice word set to obtain a branched associated customer voice word set;

[0055] Sort the branched associated customer voice word set to obtain a branched associated voice word order;

[0056] Construct a branched associated tree node sequence according to the branched associated voice word order;

[0057] Connect the branched associated tree node sequence to the last intersection tree node to obtain multiple branched tree node sequences.

[0058] Optionally, the constructing a branched associated tree node sequence according to the branched associated voice word order includes:

[0059] Successively extract branched associated voice words in the branched associated voice word order;

[0060] On the last intersection tree node, use the branched associated voice word to construct an iterative associated tree node, where the node distance between the iterative associated tree node and the last intersection tree node is the node spacing;

[0061] Determine whether the branched associated voice word is the last branched associated voice word in the branched associated voice word order;

[0062] If the branched associated voice word is not the last branched associated voice word in the branched associated voice word order, update the last intersection tree node with the iterative associated tree node, and return to the step of successively extracting branched associated voice words in the branched associated voice word order;

[0063] If the branched associated voice word is the last branched associated voice word in the branched associated voice word order, obtain a branched associated tree node sequence.

[0064] Optionally, the identifying the set of overlapping tree nodes of the currently acquired current customer voice word set and each customer voice word tree in the customer voice word tree forest includes:

[0065] Determine whether the customer voice word tree contains one or more current customer voice words in the current customer voice word set;

[0066] If the customer voice word tree does not contain one or more current customer voice words in the current customer voice word set, set the set of overlapping tree nodes to an empty set;

[0067] If the customer voice word tree contains one or more current customer voice words in the current customer voice word set, identify the overlapping tree nodes corresponding to the one or more current customer voice words in the customer voice word tree to obtain a set of overlapping tree nodes.

[0068] Optionally, before the deep learning recognition of the voice customer service intention based on the coincidence tree node set, the method further includes:

[0069] Obtain multiple groups of training customer voice word sets, and identify the training coincidence tree node sets of the training customer voice word sets and each customer voice word tree in the customer voice word tree forest;

[0070] Connect adjacent nodes of the training coincidence tree nodes belonging to the intersection tree node sequence or the same branch tree node sequence in the training coincidence tree node set to obtain a training coincidence node graph;

[0071] Identify the associated voice customer service intention corresponding to the training customer voice word set;

[0072] Train a pre-constructed initial convolutional neural network according to the training coincidence node graph and the associated voice customer service intention until the initial convolutional neural network is completed, and obtain a target convolutional neural network.

[0073] To achieve the above object, the present invention further provides a precise recognition system for voice customer service intention based on deep learning, including:

[0074] A target customer voice word set screening module, configured to obtain a customer voice text set corresponding to a voice customer service intention, perform word segmentation processing on the customer voice text set to obtain multiple groups of initial customer voice word sets; screen the multiple groups of initial customer voice word sets according to a preset initial voice repetition word order to obtain multiple groups of target customer voice word sets;

[0075] An intersection tree node sequence construction module, configured to construct an intersection tree node sequence according to the initial voice repetition word order;

[0076] A branching tree node sequence construction module is used to identify multiple groups of intersection tree node subsequences of the intersection tree node sequence, and sequentially extract intersection tree node subsequences from the multiple groups of intersection tree node subsequences according to a preset node decreasing order; determine whether there is only one intersection tree node in the intersection tree node subsequence; if there is not only one intersection tree node in the intersection tree node subsequence, then identify multiple groups of associated customer voice word sets containing the intersection tree node subsequence from the multiple groups of target customer voice word sets; identify the last intersection tree node in the intersection tree node subsequence, and construct multiple groups of branching tree node sequences on the last intersection tree node according to the multiple groups of associated customer voice word sets; remove the multiple groups of associated customer voice word sets from the multiple groups of target customer voice word sets to obtain multiple groups of iterative customer voice word sets; update the multiple groups of target customer voice word sets with the multiple groups of iterative customer voice word sets, and return to the above step of sequentially extracting intersection tree node subsequences from the multiple groups of intersection tree node subsequences according to the preset node decreasing order; if there is only one intersection tree node in the intersection tree node subsequence, then construct multiple groups of branching tree node sequences on the last intersection tree node according to the multiple groups of iterative customer voice word sets to obtain a customer voice word tree.

[0077] A voice customer service intention recognition module is used to gather the customer voice word trees corresponding to each voice customer service intention to obtain a customer voice word tree forest; identify the overlapping tree node sets between the currently acquired customer voice word set and each customer voice word tree in the customer voice word tree forest; perform deep learning recognition of the voice customer service intention according to the overlapping tree node sets to obtain the current customer service intention, where the deep learning recognition is based on a pre-trained target convolutional neural network.

[0078] To solve the above problems, the present invention also provides an electronic device, and the electronic device includes:

[0079] A memory that stores at least one instruction; and a processor that executes the instruction stored in the memory to implement the above-mentioned method for accurately identifying voice customer service intentions based on deep learning.

[0080] To solve the above problems, the present invention also provides a computer-readable storage medium, and at least one instruction is stored in the computer-readable storage medium, and the at least one instruction is executed by a processor in an electronic device to implement the above-mentioned method for accurately identifying voice customer service intentions based on deep learning.

[0081] To solve the problems described in the background art, in order to identify the user intent of the customer voice text, it is necessary to convert the customer voice text into a structured customer voice word tree. Before converting it into a customer voice word tree, it is necessary to first obtain the customer voice text set corresponding to the voice customer intent and perform word segmentation on the customer voice text set to obtain multiple groups of initial customer voice word sets. Since the customer voice word tree is constructed based on the initial voice repetition word order, and the initial customer voice word sets may not contain the initial voice repetition words, it is necessary to screen the multiple groups of initial customer voice word sets according to the preset initial voice repetition word order to obtain multiple groups of target customer voice word sets. Since the customer voice word tree contains an intersection tree node sequence and a branch tree node sequence, in order to construct the customer voice word tree, it is necessary to first construct the intersection tree node sequence, which can be constructed according to the initial voice repetition word order. Then, construct the branch tree node sequence. When constructing the branch tree node sequence, since the branch tree node sequence is connected to each intersection tree node, it is necessary to first identify multiple groups of intersection tree node subsequences of the intersection tree node sequence, and then extract the intersection tree node subsequences in turn from the multiple groups of intersection tree node subsequences according to the preset node decreasing order. At this time, the last intersection tree node of each intersection tree node subsequence corresponds to an intersection tree node that can be connected to the branch tree node sequence. Since when there is only one intersection tree node in the intersection tree node subsequence, it indicates that after connecting the branch tree node sequence to this intersection tree node, the construction of the customer voice word tree can be completed, it is necessary to first determine whether there is only one intersection tree node in the intersection tree node subsequence. If there is more than one intersection tree node in the intersection tree node subsequence, then identify multiple groups of associated customer voice word sets that contain the intersection tree node subsequence in the multiple groups of target customer voice word sets, and then identify the last intersection tree node in the intersection tree node subsequence, and construct multiple groups of branch tree node sequences on the last intersection tree node according to the multiple groups of associated customer voice word sets. Since the multiple groups of associated customer voice word sets have constructed multiple groups of branch tree node sequences, it is necessary to remove the multiple groups of associated customer voice word sets from the multiple groups of target customer voice word sets to obtain multiple groups of iterative customer voice word sets. At this time, it is necessary to update the multiple groups of target customer voice word sets with the multiple groups of iterative customer voice word sets, and re-extract the intersection tree node subsequences in turn from the multiple groups of intersection tree node subsequences according to the preset node decreasing order, and construct multiple groups of branch tree node sequences. If there is only one intersection tree node in the intersection tree node subsequence, then construct multiple groups of branch tree node sequences on the last intersection tree node according to the multiple groups of iterative customer voice word sets to obtain the customer voice word tree. Since each voice customer intent corresponds to a customer voice word tree, it is possible to collect the customer voice word trees corresponding to each voice customer intent to obtain a customer voice word tree forest. Since the greater the structural overlap between the current customer voice word set and the customer voice word tree, the closer the current customer intent is to the voice customer intent corresponding to the customer voice word tree, therefore,First, identify the overlapping tree node sets between the pre-acquired current customer voice word set and each customer voice word tree in the customer voice word tree forest, and then perform in-depth learning recognition of the voice customer service intention based on the overlapping tree node sets to obtain the current customer service intention. Therefore, the present invention can improve the recognition accuracy and recognition speed of the current intelligent customer service in voice customer service intention recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 It is a schematic flowchart of a method for accurately identifying voice customer service intention based on deep learning provided by an embodiment of the present invention;

[0083] Figure 2 It is a structural diagram of a customer voice word tree provided by an embodiment of the present invention;

[0084] Figure 3 It is a functional module diagram of a system for accurately identifying voice customer service intention based on deep learning provided by an embodiment of the present invention;

[0085] Figure 4 It is a schematic structural diagram of an electronic device for implementing the method for accurately identifying voice customer service intention based on deep learning provided by an embodiment of the present invention.

[0086] DESCRIPTION OF THE REFERENCE NUMERALS:

[0087] 1. Electronic device; 10. Processor; 11. Memory; 12. Bus.

[0088] The realization, functional features and advantages of the object of the present invention will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0089] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0090] An embodiment of the present application provides a method for accurately identifying voice customer service intention based on deep learning. The execution subject of the method for accurately identifying voice customer service intention based on deep learning includes, but is not limited to, at least one of electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the method for accurately identifying voice customer service intention based on deep learning can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to: a single server, a server cluster, a cloud server, or a cloud server cluster, etc.

[0091] Referring to Figure 1 as shown, it is a schematic flowchart of a method for accurately identifying voice customer service intention based on deep learning provided by an embodiment of the present invention. In this embodiment, the method for accurately identifying voice customer service intention based on deep learning includes:

[0092] S1. Obtain the customer voice text set corresponding to the voice customer intention, perform word segmentation on the customer voice text set, and obtain multiple groups of initial customer voice word sets.

[0093] Interpretably, the voice customer intention refers to the call intention of the user during the conversation between the customer service and the user. For example, during a power service call, the voice customer intention can be consulting package content, querying bill details, consulting business handling progress, etc. The customer voice text set refers to the text set converted from the customer voice records during the voice communication between the customer and the customer service. The word segmentation process refers to segmenting the customer voice text in the customer voice text set. The word segmentation process is a prior art and will not be elaborated here.

[0094] Furthermore, the initial customer voice word set refers to the set of initial customer voice words obtained after word segmentation of the customer voice text. Among them, each customer voice text corresponds to a group of initial customer voice word sets.

[0095] Illustratively, when the customer voice text is: "I want to change the package, but I'm not sure which one is suitable. I currently use about 60G of data per month, talk for about 300 minutes, travel frequently on business, need national roaming, and occasionally use international long-distance calls. How much does the international long-distance call cost? I may make 100 minutes of calls to the United States per month", then the initial customer voice word set can be: I, package, determine, suitable, currently, per month, data, 60G, about, talk, 300, minutes, travel on business, national roaming, occasionally, international long-distance call, charge, may, make, 100 minutes, United States.

[0096] S2. Screen the multiple groups of initial customer voice word sets according to the preset initial voice repeated word order to obtain multiple groups of target customer voice word sets.

[0097] Furthermore, the initial voice repeated word order refers to the sequence composed of the initial customer voice words that repeatedly appear in multiple groups of initial customer voice word sets. The target customer voice word set refers to the initial customer voice word set that contains one or more initial voice repeated words in the initial voice repeated word order.

[0098] In the embodiment of the present invention, before screening the multiple groups of initial customer voice word sets according to the preset initial voice repeated word order to obtain multiple groups of target customer voice word sets, the method further includes:

[0099] Obtain the customer voice union word set of the multiple groups of initial customer voice word sets, where the customer voice union word set refers to the union of each group of initial customer voice word sets;

[0100] Successively extract customer voice union words in the customer voice union word set;

[0101] Count the number of sets of the initial customer voice word sets that contain the customer voice union words in the multiple groups of initial customer voice word sets;

[0102] Determine whether the number of sets of words is greater than 1;

[0103] If the number of sets of words is not greater than 1, return the above steps of sequentially extracting the customer voice union words in the customer voice union word set;

[0104] If the number of sets of words is greater than 1, use the customer voice union words as initial voice repeated words to obtain an initial voice repeated word set;

[0105] Sort the initial voice repeated word set according to the number of sets of words of each initial voice repeated word in the initial voice repeated word set to obtain an initial voice repeated word order, where the initial voice repeated word order is sorted in descending order according to the number of sets of words.

[0106] It can be understood that the number of sets of words refers to the number of sets of the initial customer voice word sets that contain the customer voice union words. When the multiple groups of initial customer voice word sets are: I, package, confirm, suitable, monthly, traffic, about 60G; I, now, package, change to, more traffic; I, want to, activate, international roaming, prepare, what; this, video ringtone, automatic renewal, how to, cancel; then the customer voice union word set is: I, package, confirm, suitable, monthly, traffic, about 60G, now, change to, more, want to, activate, international roaming, prepare, what, this, video ringtone, automatic renewal, how to, cancel. When the customer voice union word is "traffic", the number of sets of words is 2, and at this time, "traffic" is an initial voice repeated word. The initial voice repeated word refers to the initial customer voice word that appears repeatedly in the multiple groups of initial customer voice word sets.

[0107] Furthermore, the initial voice repeated word order refers to the sequence composed of the initial voice repeated words obtained after sorting the initial voice repeated words in descending order according to the number of sets of words. For example: when the initial voice repeated words are: I, package, traffic, and the number of sets of words of the initial voice repeated word "I" is 3, the number of sets of words of "package" is 2, and the number of sets of words of "traffic" is 2, then the initial voice repeated word order is "I, package, traffic".

[0108] In the embodiment of the present invention, the screening of the multiple groups of initial customer voice word sets according to the preset initial voice repeated word order to obtain multiple groups of target customer voice word sets includes:

[0109] Sequentially extract the initial customer voice word sets in the multiple groups of initial customer voice word sets;

[0110] Determine whether the initial customer voice word set contains one or more initial voice repeated words in the initial voice repeated word order;

[0111] If the initial customer voice word set does not contain one or more initial voice repeated words in the initial voice repeated word order, return the above steps of sequentially extracting the initial customer voice word set from the multiple groups of initial customer voice word sets;

[0112] If the initial customer voice word set contains one or more initial voice repeated words in the initial voice repeated word order, use the initial customer voice word set as the target customer voice word set;

[0113] Determine whether the extraction of the initial customer voice word set is completed;

[0114] If the extraction of the initial customer voice word set is not completed, return the above steps of sequentially extracting the initial customer voice word set from the multiple groups of initial customer voice word sets;

[0115] If the extraction of the initial customer voice word set is completed, obtain multiple groups of target customer voice word sets.

[0116] Further, when the initial voice repeated word order is "I, package, traffic", the target customer voice word set can be: I, package, determine, suitable, monthly, traffic, 60G, or so; I, now, package, change to, more traffic; I, want to, open, international roaming, prepare, what, where the first target customer voice word set includes the initial voice repeated words: I, package, traffic, the second target customer voice word set includes the initial voice repeated words: I, package, traffic, and the third target customer voice word set includes the initial voice repeated word: I.

[0117] S3. Construct an intersection tree node sequence according to the initial voice repeated word order.

[0118] It can be understood that the intersection tree node sequence refers to a tree node sequence constructed according to the initial voice repeated words corresponding to the intersection of each target customer voice word set.

[0119] In the embodiment of the present invention, the constructing an intersection tree node sequence according to the initial voice repeated word order includes:

[0120] Sequentially extract the initial voice repeated words in the initial voice repeated word order;

[0121] On a preset root node, use the initial voice repeated word to construct an iterative intersection tree node, where the node distance between the iterative intersection tree node and the root node is a preset node spacing;

[0122] Determine whether the initial voice repeated word is the last initial voice repeated word in the initial voice repeated word order;

[0123] If the initial voice repeated word is not the last initial voice repeated word in the initial voice repeated word order, update the root node using the iterative intersection tree node, and return the step of sequentially extracting the initial voice repeated words in the initial voice repeated word order;

[0124] If the initial voice repeated word is the last initial voice repeated word in the initial voice repeated word order, obtain the intersection tree node sequence.

[0125] It can be understood that the root node refers to the root node of the intersection tree node sequence, and the iterative intersection tree node refers to the iteratively updated intersection tree node. The node spacing refers to the distance between adjacent nodes, which can be 2 cm.

[0126] For example, refer to Figure 2 As shown, when the initial voice repeated word order is "I, package, traffic", the intersection tree node sequence can be: I - package - traffic.

[0127] S4. Identify multiple groups of intersection tree node subsequences of the intersection tree node sequence.

[0128] It can be understood that the intersection tree node subsequence refers to a sequence composed of one or more intersection tree nodes in the intersection tree node sequence. For example: when the intersection tree node sequence is "I - package - traffic", the intersection tree node subsequence can be "I - package - traffic", "I - package", "I".

[0129] In the embodiment of the present invention, the identifying multiple groups of intersection tree node subsequences of the intersection tree node sequence includes:

[0130] Obtain the initial node number;

[0131] Determine the initial grouped tree node in the intersection tree node sequence according to the initial node number, where the initial node number is 2;

[0132] Identify the pre - tree node sequence of the initial grouped tree node, where the pre - tree node sequence refers to the tree node sequence composed of the intersection tree nodes in the intersection tree node sequence that are in front of the initial grouped tree node;

[0133] Take the pre - tree node sequence as the intersection tree node subsequence;

[0134] Perform superposition update on the initial node number, where the superposition update means adding one to the initial node number;

[0135] Obtain the total number of intersection tree nodes in the intersection tree node sequence, and calculate the target tree node serial number according to the total number of intersection tree nodes, where the difference between the target tree node serial number and the total number of intersection tree nodes is 1 and the target tree node serial number is greater than the total number of intersection tree nodes;

[0136] Determine whether the iteration node serial number is equal to the target tree node serial number;

[0137] If the iteration node serial number is not equal to the target tree node serial number, then return the step of determining the initial grouped tree node in the intersection tree node sequence according to the initial node serial number as described above;

[0138] If the iteration node serial number is equal to the target tree node serial number, then collect all the intersection tree node subsequences to obtain multiple groups of intersection tree node subsequences.

[0139] Further, the initial grouped tree node refers to the intersection tree node with the serial number of the initial node in the intersection tree node sequence. For example, when the intersection tree node sequence is: me - package - traffic, and the initial node serial number is 2, the initial grouped tree node is "package", and the pre - tree node sequence is: me. When the initial node serial number is 3, the initial grouped tree node is "traffic", and the pre - tree node sequence is: me - package. The total number of intersection tree nodes refers to the total number of intersection tree nodes in the intersection tree node sequence. For example, when the intersection tree node sequence is "me - package - traffic", the total number of intersection tree nodes is 3, and the target tree node serial number is 4.

[0140] S5. Extract the intersection tree node subsequences in multiple groups of intersection tree node subsequences in turn according to the preset node decreasing order.

[0141] Further, the node decreasing order means that the number of initial speech words in the intersection tree node subsequences decreases in turn. For example, when the multiple groups of intersection tree node subsequences are: "me - package - traffic, me - package, me", the number of initial speech words in "me - package - traffic" is 3, the number of initial speech words in "me - package" is 2, and the number of initial speech words in "me" is 1. Therefore, "me - package - traffic" is extracted first, then "me - package", and finally "me".

[0142] S6. Determine whether there is only one intersection tree node in the intersection tree node subsequence.

[0143] It is understandable that when the intersection tree node sequence is "me - package - traffic" and the intersection tree node subsequence is "me", the intersection tree node subsequence has only one intersection tree node.

[0144] If there is more than one intersection tree node in the intersection tree node subsequence, then execute S7 to identify multiple groups of associated customer speech word sets that contain the intersection tree node subsequence in multiple groups of target customer speech word sets.

[0145] It can be understood that the associated customer speech word set refers to the target customer speech word set that contains the target customer speech word order corresponding to the intersection tree node subsequence, and the target customer speech word order refers to the sequence composed of target customer speech words.

[0146] S8. Identify the last intersection tree node in the intersection tree node subsequence, and construct multiple branched tree node sequences on the last intersection tree node according to multiple groups of associated customer speech word sets.

[0147] Furthermore, the last intersection tree node refers to the last intersection tree node in the intersection tree node subsequence. The branched tree node sequence refers to the tree node sequence extended from the last intersection tree node according to the associated customer speech word set. For example: when the intersection tree node subsequence is: I - package - traffic, the associated customer speech word set can be: I, package, determine, suitable, monthly, traffic, 60G, about, and at this time, the branched tree node sequence can be: determine - suitable - monthly - 60G - about.

[0148] In the embodiment of the present invention, the constructing multiple branched tree node sequences on the last intersection tree node according to multiple groups of associated customer speech word sets includes:

[0149] Identify the target customer speech word order corresponding to the intersection tree node subsequence;

[0150] Remove the target customer speech word order from the associated customer speech word set to obtain a branched associated customer speech word set;

[0151] Sort the branched associated customer speech word set to obtain a branched associated speech word order;

[0152] Construct a branched associated tree node sequence according to the branched associated speech word order;

[0153] Connect the branched associated tree node sequence to the last intersection tree node to obtain multiple branched tree node sequences.

[0154] It is understandable that the target customer speech word order refers to the sequence composed of the target customer speech words corresponding to the intersection tree node subsequence. For example, when the intersection tree node subsequence is: I - package - traffic, the target customer speech word order is: I, package, traffic. When the associated customer speech word set is: I, package, determine, suitable, monthly, traffic, 60G, around, and the target customer speech word order is: "I, package, traffic", the branched associated customer speech word set is: determine, suitable, monthly, 60G, around. The branched associated speech word order refers to the sequence composed of the branched associated customer speech words obtained after sorting the branched associated customer speech word set. For example, the branched associated speech word order can be: determine, suitable, monthly, 60G, around. The branched associated tree node sequence refers to the tree node sequence formed by connecting the branched associated speech words in the branched associated speech word order as branched associated tree nodes using the node distance. For example, the branched associated tree node sequence can be: determine - suitable - monthly - 60G - around. Since the last intersection tree node is "traffic", "determine - suitable - monthly - 60G - around" can be connected to the last intersection tree node: "traffic" in "I - package - traffic". At this time, "determine - suitable - monthly - 60G - around" is the branched tree node sequence.

[0155] Further, referring to Figure 2 As shown, when the intersection tree node subsequence is "I - package - traffic", the last intersection tree node is the tree node corresponding to "traffic", and the multi - component branched tree node sequence can be "now - change to - more", "determine - suitable - monthly - 60G - around".

[0156] In the embodiment of the present invention, constructing the branched associated tree node sequence according to the branched associated speech word order includes:

[0157] Successively extract the branched associated speech words in the branched associated speech word order;

[0158] On the last intersection tree node, construct an iterative associated tree node using the branched associated speech word, where the node distance between the iterative associated tree node and the last intersection tree node is the node spacing;

[0159] Judge whether the branched associated speech word is the last branched associated speech word in the branched associated speech word order;

[0160] If the branched associated speech word is not the last branched associated speech word in the branched associated speech word order, update the last intersection tree node using the iterative associated tree node, and return to the above step of successively extracting the branched associated speech words in the branched associated speech word order;

[0161] If the branched associated speech word is the last branched associated speech word in the sequence of branched associated speech words, a sequence of branched associated tree nodes is obtained.

[0162] Further, the iterative associated tree node refers to a tree node that needs to be iteratively updated and represents a branched associated speech word.

[0163] S9. Exclude multiple sets of associated customer speech word sets from multiple sets of target customer speech word sets to obtain multiple sets of iterative customer speech word sets.

[0164] Understandably, the iterative customer speech word sets refer to the customer speech word sets updated after completing the construction of a sequence of multiple branched tree nodes for multiple sets of target customer speech word sets.

[0165] S10. Update multiple sets of target customer speech word sets using multiple sets of iterative customer speech word sets.

[0166] Understandably, when the multiple sets of associated customer speech word sets complete the construction of the sequence of multiple branched tree nodes, the multiple sets of associated customer speech word sets can be excluded from the sequence of multiple branched tree nodes for the next cycle of constructing the sequence of multiple branched tree nodes. Therefore, multiple sets of target customer speech word sets need to be replaced with multiple sets of iterative customer speech word sets.

[0167] Return to the above step of sequentially extracting the intersection tree node subsequences from the multiple sets of intersection tree node subsequences according to the preset decreasing order of nodes.

[0168] If there is only one intersection tree node in the intersection tree node subsequence, execute S11. Construct a sequence of multiple branched tree nodes on the last intersection tree node according to multiple sets of iterative customer speech word sets to obtain a customer speech word tree.

[0169] Understandably, when there is only one intersection tree node in the intersection tree node subsequence, the intersection tree node subsequence can be "I". At this time, each set of iterative customer speech word sets contains "I" and does not contain the initial speech repetition words corresponding to other intersection tree nodes in the intersection tree node sequence. For example, it does not include: "package", "traffic". The customer speech word tree can be referred to Figure 2 as shown.

[0170] S12. Aggregate the customer speech word trees corresponding to each voice customer intention to obtain a customer speech word tree forest.

[0171] Understandably, the customer speech word tree forest refers to a word tree forest composed of the customer speech word trees corresponding to each voice customer intention.

[0172] S13. Identify the overlapping tree node sets between the currently pre-obtained customer speech word set and each customer speech word tree in the customer speech word tree forest.

[0173] It should be understood that the current customer voice word set refers to the customer voice word set obtained after performing word segmentation on the customer voice text set of the current customer. The overlapping tree node set refers to the set of tree nodes corresponding to each current customer voice word in the current customer voice word set on the customer voice word tree, and the target customer voice word represented by the corresponding tree node is the same as the current customer voice word.

[0174] In an embodiment of the present invention, identifying the overlapping tree node set of the pre-obtained current customer voice word set and each customer voice word tree in the customer voice word tree forest includes:

[0175] Determine whether one or more current customer voice words in the current customer voice word set are included in the customer voice word tree;

[0176] If one or more current customer voice words in the current customer voice word set are not included in the customer voice word tree, set the overlapping tree node set to an empty set;

[0177] If one or more current customer voice words in the current customer voice word set are included in the customer voice word tree, identify the overlapping tree nodes corresponding to the one or more current customer voice words in the customer voice word tree to obtain an overlapping tree node set.

[0178] Further, the overlapping tree node refers to the tree node corresponding to the current customer voice word in the customer voice word tree.

[0179] S14. Perform deep learning recognition of the voice customer service intention based on the overlapping tree node set to obtain the current customer service intention, where the deep learning recognition is based on a pre-trained target convolutional neural network.

[0180] It is understandable that the current customer service intention refers to the call intention of the current customer.

[0181] In an embodiment of the present invention, before performing deep learning recognition of the voice customer service intention based on the overlapping tree node set, the method further includes:

[0182] Obtain multiple groups of training customer voice word sets, and identify the training overlapping tree node sets of the training customer voice word sets and each customer voice word tree in the customer voice word tree forest;

[0183] Connect adjacent nodes of the training overlapping tree nodes belonging to the intersection tree node sequence or the same branch tree node sequence in the training overlapping tree node set to obtain a training overlapping node graph;

[0184] Identify the associated voice customer service intention corresponding to the training customer voice word set;

[0185] Train the pre - constructed initial convolutional neural network according to the training coincidence node graph and the associated voice customer service intention until the initial convolutional neural network completes training, obtaining a target convolutional neural network.

[0186] It should be understood that the training customer voice word set refers to the set of customer voice words used to train the initial convolutional neural network, and each group of training customer voice word sets corresponds to a voice customer service intention. The training coincidence tree node set refers to the set of tree nodes corresponding to each training customer voice word in the customer voice word tree in the training customer voice word set. Since there are multiple connection methods for the training coincidence tree node set, in order to determine the unique connection method of the training coincidence tree node set, the training coincidence tree nodes belonging to the intersection tree node sequence or the same branch tree node sequence can be connected as adjacent nodes. The adjacent node connection means connecting the training coincidence tree nodes belonging to the intersection tree node sequence or the same branch tree node sequence with a distance of the node spacing. The training coincidence node graph refers to the node connection graph obtained after the adjacent node connection of the training coincidence tree node set. The associated voice customer service intention refers to the voice customer service intention corresponding to the training customer voice word set.

[0187] It can be understood that the initial convolutional neural network refers to a convolutional neural network that has not completed training. Training the initial convolutional neural network using the node connection graph and the corresponding associated voice customer service intention is prior art and will not be elaborated here. The target convolutional neural network refers to a convolutional neural network that has completed training and can recognize the corresponding associated voice customer service intention according to the node connection graph.

[0188] To solve the problems described in the background art, in order to identify the user intention of the customer voice text, it is necessary to convert the customer voice text into a structured customer voice word tree. Before converting it into a customer voice word tree, it is necessary to first obtain the customer voice text set corresponding to the voice customer intention and perform word segmentation on the customer voice text set to obtain multiple groups of initial customer voice word sets. Since the customer voice word tree is constructed based on the initial voice repetition word order, and the initial customer voice word sets may not contain the initial voice repetition words, therefore, it is necessary to screen the multiple groups of initial customer voice word sets according to the preset initial voice repetition word order to obtain multiple groups of target customer voice word sets. Since the customer voice word tree contains an intersection tree node sequence and a branch tree node sequence, therefore, in order to construct the customer voice word tree, it is necessary to first construct the intersection tree node sequence, and the intersection tree node sequence can be constructed according to the initial voice repetition word order. Then, construct the branch tree node sequence. When constructing the branch tree node sequence, since the branch tree node sequence is connected to each intersection tree node, it is necessary to first identify multiple groups of intersection tree node subsequences of the intersection tree node sequence, and then extract the intersection tree node subsequences in turn from the multiple groups of intersection tree node subsequences according to the preset node decreasing order. At this time, the last intersection tree node of each intersection tree node subsequence corresponds to an intersection tree node to which the branch tree node sequence can be connected. Since when there is only one intersection tree node in the intersection tree node subsequence, it means that after connecting the branch tree node sequence to this intersection tree node, the construction of the customer voice word tree can be completed, therefore, it is necessary to first judge whether there is only one intersection tree node in the intersection tree node subsequence. If there is not only one intersection tree node in the intersection tree node subsequence, then identify multiple groups of associated customer voice word sets that contain the intersection tree node subsequence in the multiple groups of target customer voice word sets, and then identify the last intersection tree node in the intersection tree node subsequence, and construct multiple groups of branch tree node sequences on the last intersection tree node according to the multiple groups of associated customer voice word sets. Since the multiple groups of associated customer voice word sets have constructed multiple groups of branch tree node sequences, it is necessary to remove the multiple groups of associated customer voice word sets from the multiple groups of target customer voice word sets to obtain multiple groups of iterative customer voice word sets. At this time, it is necessary to update the multiple groups of target customer voice word sets with the multiple groups of iterative customer voice word sets, and re-extract the intersection tree node subsequences in turn from the multiple groups of intersection tree node subsequences according to the preset node decreasing order, and construct multiple groups of branch tree node sequences. If there is only one intersection tree node in the intersection tree node subsequence, then construct multiple groups of branch tree node sequences on the last intersection tree node according to the multiple groups of iterative customer voice word sets to obtain the customer voice word tree. Since each voice customer intention corresponds to a customer voice word tree, it is possible to gather the customer voice word trees corresponding to each voice customer intention to obtain a customer voice word tree forest. Since the greater the structural coincidence between the current customer voice word set and the customer voice word tree, the closer the current customer intention is to the voice customer intention corresponding to the customer voice word tree, therefore,First, identify the overlapping tree node sets of the pre-acquired current customer voice word set and each customer voice word tree in the customer voice word tree forest, and then perform in-depth learning recognition of the voice customer service intention based on the overlapping tree node sets to obtain the current customer service intention. Therefore, the present invention can improve the recognition accuracy and recognition speed of the current intelligent customer service in voice customer service intention recognition.

[0189] As Figure 3 shown, it is a functional module diagram of a voice customer service intention accurate recognition system based on deep learning provided by an embodiment of the present invention.

[0190] The voice customer service intention accurate recognition system 100 based on deep learning according to the present invention can be installed in an electronic device. According to the implemented functions, the voice customer service intention accurate recognition system 100 based on deep learning can include a target customer voice word set screening module 101, an intersection tree node sequence construction module 102, a branch tree node sequence construction module 103, and a voice customer service intention recognition module 104. The modules of the present invention can also be referred to as units, which refer to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, and are stored in the memory of the electronic device.

[0191] The target customer voice word set screening module 101 is used to obtain the customer voice text set corresponding to the voice customer service intention, perform word segmentation processing on the customer voice text set to obtain multiple groups of initial customer voice word sets; and screen the multiple groups of initial customer voice word sets according to the preset initial voice repetition word order to obtain multiple groups of target customer voice word sets;

[0192] The intersection tree node sequence construction module 102 is used to construct an intersection tree node sequence according to the initial voice repetition word order;

[0193] The branch tree node sequence construction module 103 is configured to identify multiple groups of intersection tree node subsequences of the intersection tree node sequence, and sequentially extract the intersection tree node subsequences from the multiple groups of intersection tree node subsequences according to a preset decreasing order of nodes; determine whether there is only one intersection tree node in the intersection tree node subsequence; if there is not only one intersection tree node in the intersection tree node subsequence, then identify multiple groups of associated customer voice word sets including the intersection tree node subsequence from the multiple groups of target customer voice word sets; identify the last intersection tree node in the intersection tree node subsequence, and construct multiple groups of branch tree node sequences on the last intersection tree node according to the multiple groups of associated customer voice word sets; remove the multiple groups of associated customer voice word sets from the multiple groups of target customer voice word sets to obtain multiple groups of iterative customer voice word sets; update the multiple groups of target customer voice word sets with the multiple groups of iterative customer voice word sets, and return to the above step of sequentially extracting the intersection tree node subsequences from the multiple groups of intersection tree node subsequences according to the preset decreasing order of nodes; if there is only one intersection tree node in the intersection tree node subsequence, then construct multiple groups of branch tree node sequences on the last intersection tree node according to the multiple groups of iterative customer voice word sets to obtain a customer voice word tree.

[0194] The voice customer service intention recognition module 104 is configured to collect the customer voice word trees corresponding to each voice customer service intention to obtain a customer voice word tree forest; identify the overlapping tree node sets between the currently acquired customer voice word set and each customer voice word tree in the customer voice word tree forest; perform deep learning recognition of the voice customer service intention according to the overlapping tree node sets to obtain the current customer service intention, where the deep learning recognition is based on a pre-trained target convolutional neural network.

[0195] Specifically, each module in the voice customer service intention accurate recognition system 100 based on deep learning in the embodiments of the present invention uses the same technical means as those Figure 1 described in the above-mentioned voice customer service intention accurate recognition method based on deep learning, and can produce the same technical effects, which will not be elaborated here.

[0196] As Figure 4 shown, it is a schematic structural diagram of an electronic device for implementing the voice customer service intention accurate recognition method based on deep learning provided by an embodiment of the present invention.

[0197] The electronic device 1 may include a processor 10, a memory 11, and a bus 12, and may further include a computer program stored in the memory 11 and executable on the processor 10, such as a voice customer service intention accurate recognition method program based on deep learning.

[0198] Among them, the memory 11 at least includes one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (such as SD or DX memory, etc.), magnetic memory, magnetic disk, optical disc, etc. The memory 11 can be an internal storage unit of the electronic device 1 in some embodiments, such as the mobile hard disk of the electronic device 1. The memory 11 can also be an external storage device of the electronic device 1 in some other embodiments, such as a plug-in mobile hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. equipped on the electronic device 1. Further, the memory 11 also includes the internal storage unit of the electronic device 1 and also includes an external storage device. The memory 11 can be used not only to store application software installed on the electronic device 1 and various types of data, such as the code of the method program for accurate identification of voice customer service intent based on deep learning, etc., but also to temporarily store data that has been output or will be output.

[0199] The processor 10 can be composed of integrated circuits in some embodiments. For example, it can be composed of a single packaged integrated circuit, or can also be composed of multiple integrated circuits with the same or different functions packaged, including a combination of one or more Central Processing Units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, connecting all components of the entire electronic device through various interfaces and lines, and by running or executing programs or modules stored in the memory 11 (such as the method program for accurate identification of voice customer service intent based on deep learning, etc.), and calling data stored in the memory 11, to perform various functions of the electronic device 1 and process data.

[0200] The bus 12 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus 12 can be divided into an address bus, a data bus, a control bus, etc. The bus 12 is set to realize the connection and communication between the memory 11 and at least one processor 10, etc.

[0201] Figure 4 Only the electronic device with components is shown. Those skilled in the art can understand that, Figure 4The shown structure does not constitute a limitation on the electronic device 1, and it may include fewer or more components than those shown, or combine certain components, or have different component arrangements.

[0202] For example, although not shown, the electronic device 1 may further include a power source (such as a battery) for supplying power to each component. Preferably, the power source can be logically connected to the at least one processor 10 through a power management device, so as to implement functions such as charging management, discharging management, and power consumption management through the power management device. The power source may also include any components such as one or more DC or AC power sources, a recharge device, a power failure detection circuit, a power converter or inverter, and a power status indicator. The electronic device 1 may also include various sensors, a Bluetooth module, a Wi-Fi module, etc., which will not be elaborated here.

[0203] Furthermore, the electronic device 1 may further include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices.

[0204] Optionally, the electronic device 1 may further include a user interface. The user interface may be a display (Display), an input unit (such as a keyboard (Keyboard)). Optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. Among them, the display may also be appropriately referred to as a display screen or a display unit, which is used to display the information processed in the electronic device 1 and to display a visual user interface.

[0205] The program of the method for accurately identifying the intention of a voice customer service based on deep learning stored in the memory 11 in the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can implement:

[0206] Obtain a set of customer voice texts corresponding to the intention of the voice customer service, perform word segmentation processing on the set of customer voice texts to obtain multiple sets of initial customer voice word sets;

[0207] Filter the multiple sets of initial customer voice word sets according to a preset initial voice repetition word order to obtain multiple sets of target customer voice word sets;

[0208] Construct an intersection tree node sequence according to the initial voice repetition word order;

[0209] Identify multiple sets of intersection tree node subsequences of the intersection tree node sequence;

[0210] Extract the intersection tree node subsequences in turn from multiple groups of intersection tree node subsequences according to the preset decreasing order of nodes;

[0211] Determine whether there is only one intersection tree node in the intersection tree node subsequence;

[0212] If there is more than one intersection tree node in the intersection tree node subsequence, identify multiple groups of associated customer voice word sets that contain the intersection tree node subsequence in multiple groups of target customer voice word sets;

[0213] Identify the last intersection tree node in the intersection tree node subsequence, and construct multiple groups of branched tree node sequences on the last intersection tree node according to multiple groups of associated customer voice word sets;

[0214] Remove multiple groups of associated customer voice word sets from multiple groups of target customer voice word sets to obtain multiple groups of iterative customer voice word sets;

[0215] Update multiple groups of target customer voice word sets with multiple groups of iterative customer voice word sets, and return the step of extracting the intersection tree node subsequences in turn from multiple groups of intersection tree node subsequences according to the preset decreasing order of nodes;

[0216] If there is only one intersection tree node in the intersection tree node subsequence, construct multiple groups of branched tree node sequences on the last intersection tree node according to multiple groups of iterative customer voice word sets to obtain a customer voice word tree;

[0217] Collect the customer voice word trees corresponding to each voice customer intention to obtain a customer voice word tree forest;

[0218] Identify the overlapping tree node sets between the currently obtained customer voice word set and each customer voice word tree in the customer voice word tree forest;

[0219] Perform in-depth learning recognition of the voice customer intention according to the overlapping tree node sets to obtain the current customer intention, where the in-depth learning recognition is based on a pre-trained target convolutional neural network.

[0220] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figures 1 to 4 the description of the relevant steps in the corresponding embodiment, which will not be elaborated here.

[0221] Furthermore, if the modules / units integrated in the electronic device 1 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM, Read-Only Memory).

[0222] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor of an electronic device, it can implement:

[0223] Obtain a set of customer voice texts corresponding to the voice customer intent, perform word segmentation on the set of customer voice texts to obtain multiple sets of initial customer voice word sets;

[0224] Filter the multiple sets of initial customer voice word sets according to a preset initial voice repetition word order to obtain multiple sets of target customer voice word sets;

[0225] Construct an intersection tree node sequence according to the initial voice repetition word order;

[0226] Identify multiple sets of intersection tree node subsequences of the intersection tree node sequence;

[0227] Extract the intersection tree node subsequences in turn from the multiple sets of intersection tree node subsequences according to a preset node decreasing order;

[0228] Judge whether there is only one intersection tree node in the intersection tree node subsequence;

[0229] If there is not only one intersection tree node in the intersection tree node subsequence, then identify multiple sets of associated customer voice word sets that contain the intersection tree node subsequence in the multiple sets of target customer voice word sets;

[0230] Identify the last intersection tree node in the intersection tree node subsequence, and construct multiple sets of branch tree node sequences on the last intersection tree node according to the multiple sets of associated customer voice word sets;

[0231] Remove the multiple sets of associated customer voice word sets from the multiple sets of target customer voice word sets to obtain multiple sets of iterative customer voice word sets;

[0232] Update the multiple sets of target customer voice word sets with the multiple sets of iterative customer voice word sets, and return to the step of extracting the intersection tree node subsequences in turn from the multiple sets of intersection tree node subsequences according to the preset node decreasing order;

[0233] If there is only one intersection tree node in the intersection tree node subsequence, a multi-component branch tree node sequence is constructed on the last intersection tree node according to multiple groups of iterative customer voice word sets to obtain a customer voice word tree;

[0234] The customer voice word trees corresponding to each voice customer intention are assembled to obtain a customer voice word tree forest;

[0235] Identify the overlapping tree node sets of the currently pre-acquired customer voice word set and each customer voice word tree in the customer voice word tree forest;

[0236] Deep learning recognition of the voice customer intention is performed according to the overlapping tree node sets to obtain the current customer intention, where the deep learning recognition is based on a pre-trained target convolutional neural network.

[0237] In several embodiments provided by the present invention, it should be understood that the disclosed devices, systems, and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative, and there can be other partitioning methods in actual implementation.

[0238] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0239] In addition, the functional modules in each embodiment of the present invention can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated units can be implemented in the form of hardware, or in the form of a combination of hardware and software functional modules.

[0240] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and without departing from the spirit or basic characteristics of the present invention, the present invention can be implemented in other specific forms.

[0241] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for accurately identifying the intent of a voice customer service based on deep learning, characterized in that, The method includes: Obtain a set of customer voice texts corresponding to the voice customer intention, perform word segmentation on the set of customer voice texts to obtain multiple sets of initial customer voice word sets; Screen the multiple sets of initial customer voice word sets according to a preset initial voice repetition word order to obtain multiple sets of target customer voice word sets; Construct an intersection tree node sequence according to the initial voice repetition word order; Identify multiple sets of intersection tree node subsequences of the intersection tree node sequence; Extract the intersection tree node subsequences in turn from the multiple sets of intersection tree node subsequences according to a preset node decreasing order; Judge whether there is only one intersection tree node in the intersection tree node subsequence; If there is not only one intersection tree node in the intersection tree node subsequence, identify multiple sets of associated customer voice word sets containing the intersection tree node subsequence in the multiple sets of target customer voice word sets; Identify the last intersection tree node in the intersection tree node subsequence, and construct multiple sets of branched tree node sequences on the last intersection tree node according to the multiple sets of associated customer voice word sets; Exclude the multiple sets of associated customer voice word sets from the multiple sets of target customer voice word sets to obtain multiple sets of iterative customer voice word sets; Update the multiple sets of target customer voice word sets with the multiple sets of iterative customer voice word sets, and return to the step of extracting the intersection tree node subsequences in turn from the multiple sets of intersection tree node subsequences according to the preset node decreasing order; If there is only one intersection tree node in the intersection tree node subsequence, construct multiple sets of branched tree node sequences on the last intersection tree node according to the multiple sets of iterative customer voice word sets to obtain a customer voice word tree; Collect the customer voice word trees corresponding to each voice customer intention to obtain a customer voice word tree forest; Identify the set of overlapping tree nodes between the currently obtained customer voice word set and each customer voice word tree in the customer voice word tree forest; Perform deep learning recognition of the voice customer intention according to the set of overlapping tree nodes to obtain the current customer intention, where the deep learning recognition is based on a pre-trained target convolutional neural network.

2. The method for accurately identifying the intent of a voice customer service based on deep learning according to claim 1, wherein The screening of the multiple sets of initial customer voice word sets according to a preset initial voice repetition word order to obtain multiple sets of target customer voice word sets includes: Extract the initial customer voice word sets in turn from the multiple sets of initial customer voice word sets; Judge whether the initial customer voice word set contains one or more initial voice repetitions in the initial voice repetition word order; If the initial customer voice word set does not contain one or more initial voice repetitions in the initial voice repetition word order, return to the step of extracting the initial customer voice word sets in turn from the multiple sets of initial customer voice word sets; If the initial customer voice word set contains one or more initial voice repetitions in the initial voice repetition word order, use the initial customer voice word set as the target customer voice word set; Judge whether the initial customer voice word set has been extracted completely; If the initial customer voice word set has not been extracted completely, return to the step of extracting the initial customer voice word sets in turn from the multiple sets of initial customer voice word sets; If the initial customer voice word set has been extracted completely, obtain multiple sets of target customer voice word sets.

3. The method for accurately identifying the intent of a voice customer service based on deep learning according to claim 2, wherein Before screening multiple groups of initial customer voice word sets according to a preset initial voice repetition word order to obtain multiple groups of target customer voice word sets, the method further includes: Obtaining a customer voice union word set of the multiple groups of initial customer voice word sets, where the customer voice union word set refers to the union of each group of initial customer voice word sets; Sequentially extracting customer voice union words from the customer voice union word set; Counting the number of word sets in the multiple groups of initial customer voice word sets that contain the customer voice union word; Judging whether the number of word sets is greater than 1; If the number of word sets is not greater than 1, return to the step of sequentially extracting customer voice union words from the customer voice union word set; If the number of word sets is greater than 1, use the customer voice union word as an initial voice repetition word to obtain an initial voice repetition word set; Sort the initial voice repetition word set according to the number of word sets of each initial voice repetition word in the initial voice repetition word set to obtain an initial voice repetition word order, where the initial voice repetition word order is sorted in descending order according to the number of word sets.

4. The method for accurately identifying the intent of a voice customer service based on deep learning according to claim 3, wherein The constructing an intersection tree node sequence according to the initial voice repetition word order includes: Sequentially extracting initial voice repetition words from the initial voice repetition word order; On a preset root node, constructing an iterative intersection tree node using the initial voice repetition word, where the node distance between the iterative intersection tree node and the root node is a preset node spacing; Judging whether the initial voice repetition word is the last initial voice repetition word in the initial voice repetition word order; If the initial voice repetition word is not the last initial voice repetition word in the initial voice repetition word order, update the root node using the iterative intersection tree node, and return to the step of sequentially extracting initial voice repetition words from the initial voice repetition word order; If the initial voice repetition word is the last initial voice repetition word in the initial voice repetition word order, obtain an intersection tree node sequence.

5. The method for accurately identifying the intent of a voice customer service based on deep learning according to claim 4, wherein, The identifying multiple groups of intersection tree node subsequences of the intersection tree node sequence includes: Obtaining an initial node number; Determining an initial grouped tree node in the intersection tree node sequence according to the initial node number, where the initial node number is 2; Identifying a pre-tree node sequence of the initial grouped tree node, where the pre-tree node sequence refers to a tree node sequence composed of intersection tree nodes in the intersection tree node sequence that are in front of the initial grouped tree node; Using the pre-tree node sequence as an intersection tree node subsequence; Performing an overlay update on the initial node number, where the overlay update means adding one to the initial node number; Obtaining the total number of intersection tree nodes of the intersection tree node sequence, and calculating a target tree node number according to the total number of intersection tree nodes, where the difference between the target tree node number and the total number of intersection tree nodes is 1 and the target tree node number is greater than the total number of intersection tree nodes; Judging whether the iterative node number is equal to the target tree node number; If the iteration node serial number is not equal to the target tree node serial number, return the step of determining the initial grouped tree node in the intersection tree node sequence according to the initial node serial number as described above; If the iteration node serial number is equal to the target tree node serial number, collect all the intersection tree node subsequences to obtain multiple groups of intersection tree node subsequences.

6. The method for accurately identifying the intent of a voice customer service based on deep learning according to claim 5, characterized in that, The constructing of multiple branched tree node sequences on the last intersection tree node according to multiple groups of associated customer voice word sets includes: Identifying the target customer voice word order corresponding to the intersection tree node subsequence; Removing the target customer voice word order from the associated customer voice word set to obtain a branched associated customer voice word set; Sorting the branched associated customer voice word set to obtain a branched associated voice word order; Constructing a branched associated tree node sequence according to the branched associated voice word order; Connecting the branched associated tree node sequence to the last intersection tree node to obtain multiple branched tree node sequences.

7. The method for accurately identifying the intent of a voice customer service based on deep learning according to claim 6, characterized in that, The constructing of a branched associated tree node sequence according to the branched associated voice word order includes: Successively extracting branched associated voice words in the branched associated voice word order; On the last intersection tree node, constructing an iterative associated tree node using the branched associated voice word, where the node distance between the iterative associated tree node and the last intersection tree node is the node spacing; Judging whether the branched associated voice word is the last branched associated voice word in the branched associated voice word order; If the branched associated voice word is not the last branched associated voice word in the branched associated voice word order, update the last intersection tree node using the iterative associated tree node, and return the step of successively extracting branched associated voice words in the branched associated voice word order as described above; If the branched associated voice word is the last branched associated voice word in the branched associated voice word order, obtain a branched associated tree node sequence.

8. The method for accurately identifying the intent of a voice customer service based on deep learning according to claim 7, wherein The identifying of the set of overlapping tree nodes between the currently acquired current customer voice word set and each customer voice word tree in the customer voice word tree forest includes: Judging whether the customer voice word tree contains one or more current customer voice words in the current customer voice word set; If the customer voice word tree does not contain one or more current customer voice words in the current customer voice word set, set the set of overlapping tree nodes as an empty set; If the customer voice word tree contains one or more current customer voice words in the current customer voice word set, identify the overlapping tree nodes corresponding to the one or more current customer voice words in the customer voice word tree to obtain a set of overlapping tree nodes.

9. The method for accurately identifying the intent of a voice customer service based on deep learning according to claim 8, wherein Before the deep learning identification of the voice customer service intention according to the set of overlapping tree nodes, the method further includes: Obtaining multiple groups of training customer voice word sets, and identifying the set of training overlapping tree nodes between the training customer voice word sets and each customer voice word tree in the customer voice word tree forest; Connecting adjacent nodes of the training overlapping tree nodes belonging to the intersection tree node sequence or the same branched tree node sequence in the set of training overlapping tree nodes to obtain a training overlapping node graph; Identifying the associated voice customer service intention corresponding to the training customer voice word set; Train the pre-constructed initial convolutional neural network according to the training coincidence node graph and the associated voice customer service intention until the initial convolutional neural network is completed, and obtain the target convolutional neural network.

10. A voice customer service intent precise recognition system based on deep learning, characterized in that, The system includes: A target customer voice word set screening module, configured to obtain a customer voice text set corresponding to a voice customer service intention, perform word segmentation processing on the customer voice text set to obtain multiple groups of initial customer voice word sets; screen the multiple groups of initial customer voice word sets according to a preset initial voice repetition word order to obtain multiple groups of target customer voice word sets; An intersection tree node sequence construction module, configured to construct an intersection tree node sequence according to the initial voice repetition word order; A branch tree node sequence construction module, configured to identify multiple groups of intersection tree node subsequences of the intersection tree node sequence, and sequentially extract intersection tree node subsequences from the multiple groups of intersection tree node subsequences according to a preset node decreasing order; determine whether there is only one intersection tree node in the intersection tree node subsequence; if there is not only one intersection tree node in the intersection tree node subsequence, then identify multiple groups of associated customer voice word sets including the intersection tree node subsequence in the multiple groups of target customer voice word sets; identify the last intersection tree node in the intersection tree node subsequence, and construct multiple groups of branch tree node sequences on the last intersection tree node according to the multiple groups of associated customer voice word sets; remove the multiple groups of associated customer voice word sets from the multiple groups of target customer voice word sets to obtain multiple groups of iterative customer voice word sets; update the multiple groups of target customer voice word sets with the multiple groups of iterative customer voice word sets, and return to the above step of sequentially extracting intersection tree node subsequences from the multiple groups of intersection tree node subsequences according to the preset node decreasing order; if there is only one intersection tree node in the intersection tree node subsequence, then construct multiple groups of branch tree node sequences on the last intersection tree node according to the multiple groups of iterative customer voice word sets to obtain a customer voice word tree; A voice customer service intention recognition module, configured to collect customer voice word trees corresponding to each voice customer service intention to obtain a customer voice word tree forest; identify a coincidence tree node set between a currently pre-obtained customer voice word set and each customer voice word tree in the customer voice word tree forest; perform deep learning recognition of the voice customer service intention according to the coincidence tree node set to obtain the current customer service intention, where the deep learning recognition is based on the pre-trained target convolutional neural network.

Images