Information processing method and device, equipment, storage medium and computer program product

Abstract

The invention provides an information processing method and device, equipment, a storage medium and a computer program product, and the information processing method comprises the steps: determining node information of a business process node where a dialogue is located according to a dialogue text; under the condition that the node information indicates that the child node under the main process node where the dialogue is located is not ended, first candidate verbal skill information in historical process verbal skill is acquired for the child node; and determining target verbal skill information according to the first candidate verbal skill information. According to the scheme, the accurate target verbal skill can be automatically obtained, so that the problems of high implementation cost, poor experience, low accuracy and the like caused by manual participation are avoided as much as possible.

Description

Technical Field

[0001] The present application relates to the field of information processing technology, and in particular to an information processing method, apparatus, device, storage medium, and computer program product. Background Art

[0002] In telemarketing scenarios, companies typically need to invest significant resources in training human agents on sales pitches. This not only increases labor costs but also prevents agents from being available 24 / 7, impacting service quality and customer experience. Furthermore, when handling complex conversations or addressing diverse customer needs, human agents may lack sufficient flexibility and adaptability, resulting in poor sales pitches and low accuracy in their responses.

[0003] As described above, the information processing solutions for speech techniques in the existing technology have problems such as high cost and low accuracy. Summary of the Invention

[0004] The purpose of this application is to provide an information processing method, device, equipment, storage medium and computer program product to solve the problems of high cost and low accuracy in the information processing solutions for speech in the existing technology.

[0005] In order to solve the above technical problems, the present invention provides an information processing method, including:

[0006] Determine the node information of the business process node where the conversation is currently located based on the conversation text;

[0007] When the node information indicates that the child node under the main process node where the conversation is currently located has not ended, obtaining the first candidate speech information in the historical process speech for the child node;

[0008] Determine target speech information based on the first candidate speech information.

[0009] Optionally, obtaining the first candidate speech information in the historical process speech includes:

[0010] Obtain at least two candidate items corresponding to the child node from the historical process dialogue;

[0011] Obtain the second candidate speech information output by the speech model;

[0012] Obtaining similarities between the second candidate speech information and each of the candidate items;

[0013] Based on the similarity, the first candidate speech information is determined from the at least two candidate items.

[0014] Optionally, determining target speech information based on the first candidate speech information includes:

[0015] Obtaining a perplexity value between the user text and the first candidate speech information in the conversation text; the perplexity value is used to determine text parameters of the first candidate speech information, the text parameters including at least one of a fluency parameter and a rationality parameter;

[0016] Target speech information is determined based on the confusion value.

[0017] Optionally, determining target speech information according to the perplexity value includes:

[0018] Obtaining the minimum perplexity value as the target perplexity value;

[0019] When the target perplexity value satisfies the output condition, the first candidate speech information corresponding to the target perplexity value is used as the target speech information;

[0020] When the target perplexity value does not satisfy the output condition, the preset speech information is used as the target speech information.

[0021] Optionally, determining the node information of the business process node where the conversation is currently located based on the conversation text includes:

[0022] Using the recognition model, the node information of the business process node where the conversation is currently located is determined based on the conversation text.

[0023] Optionally, also include:

[0024] Using the labeling model, the historical dialogue text is labeled according to the historical dialogue data and business process information to obtain the historical process dialogue.

[0025] The present application also provides an information processing device, including:

[0026] A first determination module is used to determine the node information of the business process node where the conversation is currently located based on the conversation text;

[0027] A first acquisition module is configured to acquire, for a child node under a main process node where the conversation is currently located, first candidate speech information in historical process speech for the child node when the node information indicates that the child node has not ended;

[0028] The second determination module is used to determine target speech information based on the first candidate speech information.

[0029] Optionally, obtaining the first candidate speech information in the historical process speech includes:

[0030] Obtain at least two candidate items corresponding to the child node from the historical process dialogue;

[0031] Obtain the second candidate speech information output by the speech model;

[0032] Obtaining similarities between the second candidate speech information and each of the candidate items;

[0033] Based on the similarity, the first candidate speech information is determined from the at least two candidate items.

[0034] Optionally, determining target speech information based on the first candidate speech information includes:

[0035] Obtaining a perplexity value between the user text and the first candidate speech information in the conversation text; the perplexity value is used to determine text parameters of the first candidate speech information, the text parameters including at least one of a fluency parameter and a rationality parameter;

[0036] Target speech information is determined based on the confusion value.

[0037] Optionally, determining target speech information according to the perplexity value includes:

[0038] Obtaining the minimum perplexity value as the target perplexity value;

[0039] When the target perplexity value satisfies the output condition, the first candidate speech information corresponding to the target perplexity value is used as the target speech information;

[0040] When the target perplexity value does not satisfy the output condition, the preset speech information is used as the target speech information.

[0041] Optionally, determining the node information of the business process node where the conversation is currently located based on the conversation text includes:

[0042] Using the recognition model, the node information of the business process node where the conversation is currently located is determined based on the conversation text.

[0043] Optionally, also include:

[0044] The first labeling module is used to use the labeling model to label the historical dialogue text according to the historical dialogue data and business process information to obtain the historical process dialogue.

[0045] An embodiment of the present application also provides an information processing device, including a memory, a processor, and a program stored in the memory and executable on the processor; when the processor executes the program, the above-mentioned information processing method is implemented.

[0046] An embodiment of the present application further provides a readable storage medium on which a program is stored, and when the program is executed by a processor, the steps in the above-mentioned information processing method are implemented.

[0047] An embodiment of the present application further provides a computer program product, comprising computer instructions, which implement the steps of the above-mentioned information processing method when executed by a processor.

[0048] The beneficial effects of the above technical solution of this application are as follows:

[0049] In the above scheme, the information processing method determines the node information of the business process node where the conversation is currently located based on the conversation text; when the node information indicates that the child node under the main process node where the conversation is currently located has not ended, the first candidate speech information in the historical process speech is obtained for the child node; the target speech information is determined based on the first candidate speech information; it can support the automatic acquisition of accurate target speech, thereby minimizing problems such as high implementation cost, poor experience and low accuracy caused by manual participation. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 This is a flowchart of the information processing method according to an embodiment of the present application;

[0051] Figure 2 This is a schematic diagram of a specific implementation flow of the information processing method according to an embodiment of the present application;

[0052] Figure 3 This is a functional framework diagram of an embodiment of the present application;

[0053] Figure 4 This is a schematic diagram of the structure of an information processing device according to an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to make the technical problems, technical solutions and advantages to be solved by this application clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0055] This application aims to solve the problems of high cost and low accuracy in the information processing scheme for speech in the existing technology, and provides an information processing method, such as Figure 1 Shown, including:

[0056] Step 11: Determine the node information of the business process node where the conversation is currently located based on the conversation text;

[0057] Step 12: When the node information indicates that the child node under the main process node where the conversation is currently located has not ended, obtain the first candidate speech word information in the historical process speech words for the child node;

[0058] Step 13: Determine target speech information based on the first candidate speech information.

[0059] In this solution, after executing step 13, operations such as "providing speech reminders based on the target speech information" may also be executed, which is not limited here.

[0060] The information processing method provided in the embodiment of the present application determines the node information of the business process node where the conversation is currently located based on the conversation text; when the node information indicates that the child node under the main process node where the conversation is currently located has not ended, obtains the first candidate speech information in the historical process speech for the child node; determines the target speech information based on the first candidate speech information; and can support the automatic acquisition of accurate target speech, thereby minimizing problems such as high implementation cost, poor experience and low accuracy caused by manual participation.

[0061] The step of obtaining the first candidate speech information from the historical process speech includes: obtaining at least two candidate items (i.e., multiple candidate speech items) corresponding to the child node from the historical process speech; obtaining the second candidate speech information output by the speech model; obtaining the similarity between the second candidate speech information and each of the candidate items; and determining the first candidate speech information from the at least two candidate items based on the similarity. In this way, the first candidate speech information can be accurately obtained. The speech model can be implemented using a large model, etc., but is not limited to this.

[0062] In an embodiment of the present application, determining the target speech information based on the first candidate speech information includes: obtaining a perplexity value between the user text in the conversation text and the first candidate speech information; the perplexity value is used to determine the text parameters of the first candidate speech information, and the text parameters include: at least one of a fluency parameter and a rationality parameter; and determining the target speech information based on the perplexity value. In this way, better target speech information can be obtained. Among them, the number of first candidate speech information can be at least one, and "obtaining the perplexity value between the user text in the conversation text and the first candidate speech information" can include: obtaining the perplexity value between the user text and each first candidate speech information; and / or, the user text can be used as the upper context in the perplexity calculation, and the first candidate speech information can be used as the lower context in the perplexity calculation, but is not limited to this.

[0063] The step of determining the target speech information based on the perplexity value includes: obtaining the perplexity value with the minimum value as the target perplexity value; if the target perplexity value satisfies an output condition (e.g., is less than a threshold), using the first candidate speech information corresponding to the target perplexity value as the target speech information; and if the target perplexity value does not satisfy the output condition, using preset speech information (e.g., a preset fallback speech) as the target speech information. This can specifically implement the acquisition of the target speech information.

[0064] In the embodiments of the present application, determining the node information of the business process node at which the conversation is currently located based on the conversation text includes: utilizing a recognition model to determine the node information of the business process node at which the conversation is currently located based on the conversation text. This allows accurate acquisition of the node information. The recognition model can be implemented using a large model, for example, but is not limited thereto.

[0065] Furthermore, the information processing method further includes utilizing a tagging model to tag historical conversation text based on historical conversation data and business process information to obtain the historical process dialogue. This allows accurate acquisition of the historical process dialogue. The tagging model can be implemented using a large model, and / or the business process information can include, but is not limited to, main process information and sub-process information.

[0066] In the embodiment of the present application, the information processing method may further include:

[0067] When the node information indicates that a sub-node under the main process node where the conversation is currently located has ended, determining the next sub-node information; and providing a speech reminder based on the next sub-node information. Specifically, "providing a speech reminder based on the next sub-node information" may include: determining the speech corresponding to the next sub-node information based on a preset process speech template, and providing the speech reminder;

[0068] And / or, when the node information indicates that the main process node of the current conversation has ended, a reminder is given according to the ending dialogue, or the conversation is ended. This can support the integrity of the solution.

[0069] The information processing method provided in the embodiments of the present application is described below with examples.

[0070] In response to the above technical problems, the embodiment of the present application provides an information processing method, which can be specifically implemented as a solution for generating intelligent speech, which can support solving the following problems: 1) the cost of manual seats is high and they cannot be online all day; 2) the pre-training and fine-tuning of large models are time-consuming and labor-intensive, and cannot be implemented in projects, and the implementation cost is high; 3) the cost of retraining large models after business process changes is high; the implementation of this solution can use a large model (such as at least one of the Qwen, GPT and other models) as the basic model, without the need for pre-training and fine-tuning. Specifically, this solution can be as follows: Figure 2 As shown, including the following:

[0071] 1) Historical dialogue marking:

[0072] Based on the historical conversation data (in the conversation record library), combined with the tag tree tag_tree of the business process system (including the main process and sub-processes), the large model is used to automatically tag the agent's speech text one by one (this can avoid manual tagging work). In other words, historical conversations are tagged to obtain a process speech library (which can correspond to the above-mentioned use of the tagging model, based on historical conversation data and business process information, to tag historical speech text to obtain the historical process speech). An example of the prompt word prompt_tag involved in tagging is as follows:

[0073] prompt_tag="###Task: Please determine which main process node and sub-process node the latest conversation belongs to based on the business process system labels and conversation text.

[0074] ###Business process system:

[0075] {labels}

[0076] ###Dialogue Text:

[0077] Agent: Have you checked in yet?

[0078] ###Output process nodes in json structure:

[0079] {'main process node':'',

[0080] 'Sub-process node':''}". Specifically, it can be output according to the format format(labels=tag_tree), where the tag_tree example is as follows:

[0081] tag_tree={

[0082] 'Main process': ['Greeting', 'Add user WeChat', 'Company introduction', 'Introduction to promotions', 'Collect user information', 'Invite to the store', 'Conclusion'],

[0083] 'Subprocess':{

[0084] 'Greet':['Greet'],

[0085] 'Add user WeChat': ['Request to add WeChat', 'Add user phone', 'Provide design case'],

[0086] 'Company Introduction': ['Company Introduction'],

[0087] 'Introduction to promotional activities': ['Activity time and address', 'Introduction to promotional information'],

[0088] 'Collect user information':['Has the house been delivered', 'Ask about the delivery time', 'Has the house been decorated',

[0089] 'Ask when to renovate', 'Ask if it is the first home', 'Ask user preferences', 'Ask user budget'],

[0090] 'Invite to the store': ['Invite to the store', 'Introduction to store gifts'],

[0091] 'Conclusion':['Conclusion']}

[0092] }

[0093] For the above, the expected output result output is as follows (i.e., for the above example of "outputting process nodes in JSON structure"):

[0094] output = {'main process node': 'collect user information', 'sub-process node': 'whether the house has been delivered'}.

[0095] In addition, after all agent scripts are marked, they can be stored as a whole in the process script library. The script library can contain the agent scripts, their corresponding main process nodes, and sub-process nodes, such as the following:

[0096] [{'text':'Agent: Hello, is this Mr. Li?','Main Process Node':'Greetings','Sub-Process Node':'Greetings'},{'text':'Agent: Have you checked in yet?','Main Process Node':'Collecting User Information','Sub-Process Node':'Has the room been delivered?'}...{'text':'Agent: To facilitate follow-up service, I'd like to add you on WeChat. Is this your phone number?','Main Process Node':'Add User WeChat','Sub-Process Node':'Request to Add WeChat'},{'text':'Agent: Okay, I won't bother you anymore. Goodbye.','Main Process Node':'Conclusion','Sub-Process Node':'Conclusion'}].

[0097] 2) Read QA (question and answer) text (i.e., conversation text):

[0098] When the system is initialized, you can greet the customer first, and after the user responds, it starts to receive the conversation text between the agent and the user, that is, the question and answer pairs. For example:

[0099] text_i="Agent: Hello, I am from xx Decoration Company. Are you Ms. Li?

[0100] Customer: Yes.

[0101] Agent: Have you checked in?

[0102] Customer: Received.

[0103] Agent: When do you plan to renovate?

[0104] Client: August.

[0105] 3) Process node identification:

[0106] Integrating this with the business process, a large model (such as Qwen, GPT, etc.; corresponding to the aforementioned recognition model) can be used to analyze the QA text text_i. Specifically, by writing a prompt word prompt_main and inputting it into the large model, the model can identify the process node at which the latest conversation is currently located (this corresponds to using the recognition model to determine the node information of the business process node the conversation is currently located at based on the conversation text). However, since the customer's answer is uncertain, the process direction is uncertain. Therefore, this solution utilizes the large model to semantically understand the customer's statement and, within the constraints of the given process, leverage the large model's generation capabilities to provide the next agent's speech. For example:

[0107] prompt_main_i="###Task: Based on the business process system labels and conversation text, determine which main process node or sub-process node the current latest conversation belongs to. Also, based on the customer's response, determine whether the sub-process has ended. If the sub-process has not ended, provide the corresponding agent script; otherwise, leave the script blank.

[0108] ###Business process system:

[0109] {labels}

[0110] ###Dialogue Text:

[0111] Agent: Hello, I am from xx decoration company. Are you Ms. Li?

[0112] Customer: Yes.

[0113] Agent: Have you checked in?

[0114] Customer: Received.

[0115] Agent: When do you plan to renovate?

[0116] Client: August.

[0117] ###Output the result in json structure:

[0118] {'main process node':'',

[0119] 'Sub-process node':'',

[0120] 'Is the sub-process ended':'',

[0121] 'Agent's Talk':''}". Specifically, it can be output in the format format(labels=tag_tree),

[0122] Specifically, the expected output result output_main is as follows:

[0123] output_main={

[0124] 'Main process node': 'Collect user information',

[0125] 'Sub-process node': 'Ask when to renovate',

[0126] 'Is the sub-process ended': 'Yes',

[0127] 'Agent Talk':''}; that is, the talk output is empty.

[0128] 4) Whether to conclude:

[0129] Based on the output result output_main, determine whether the current conversation stage is the "Conclusion" node of the business process. If so, for example, if output_main is {'Main Process Node': 'Conclusion', 'Sub-Process Node': 'Conclusion', 'Is Sub-Process Ended': 'Yes', 'Agent Script': ''}, then the outbound conversation ends. If not, proceed to the next step.

[0130] 5) Whether the sub-process ends:

[0131] Perform relevant operations based on the 'Is the sub-process ended' field of the output result output_main. For example, if the value of the 'Is the sub-process ended' field indicates 'yes', proceed to the 'Generate new process script' step; otherwise, proceed to the next step of recalling similar scripts.

[0132] The above 3) mainly targets the main business process. If it is judged to be "conclusion", the entire system process is over; 4) mainly targets the sub-processes (such as: ['Request to add WeChat', 'Add user phone', 'Provide design case']) of the business process node (such as: add user WeChat). If all sub-processes are completed, then the current main process node is completed and you can enter the next main process node.

[0133] 6) Generate new process scripts:

[0134] Combine the business process and the current main process node to determine the next process node. Use the preset process speech template to give the agent speech that needs to be output at the next process node (i.e., new process speech), recorded as output_speech, and return the agent speech.

[0135] 7) Recall similar words:

[0136] When the value of the 'Is the sub-process ended' field in the output result output_main indicates 'no', for example, the latest conversation at this time is "Agent: When do you plan to renovate? Customer: I'm out of town, let's talk about it when I get back.", based on the value of the 'sub-process node' field ('Ask when the renovation will be done', recorded as child_process_now) and the 'agent dialogue' given by the large model ("When will you be back?"), the agent dialogue with the 'sub-process node' of 'Has the house been delivered' can be filtered out from the process dialogue library as a candidate, such as: [{'text':'Agent: Hello, when do you plan to renovate?', 'Main process node': 'Collect user information', 'Sub-process node': 'Ask when the renovation will be done'}, {'text':'Agent: Which month will you renovate?', 'Main process node': 'Collect user information', 'Sub-process node': 'Ask when the renovation will be done'}... {'text':'Agent: Which month will you be back? You send me a letter in advance', 'main process node': 'collect user information', 'sub-process node': 'ask when the decoration will be done'}]; then, by calculating the text similarity between the candidate and the 'agent's words' generated by the large model, the topN (N=3) most similar words can be selected as the new candidate candidate_text (corresponding to the first candidate word information mentioned above). For example, the top3 words are: candidate_text = [{'text':'Agent: When do you plan to come back? ', 'main process node': 'collect user information', 'sub-process node': 'ask when the decoration will be done'}, {'text':'Agent: Which month will you come back here? ', 'main process node': 'collect user information', 'sub-process node': 'ask when the decoration will be done'}, {'text':'Agent: When will you come back? I will make an appointment for you in advance', 'main process node': 'collect user information', 'sub-process node': 'ask when the decoration will be done'}]. This part of the content may correspond to the above-mentioned case where the node information indicates that the child node under the main process node where the conversation is currently located has not ended, and the first candidate speech information in the historical process speech is obtained for the child node; the obtaining of the first candidate speech information in the historical process speech includes: obtaining at least two candidate items corresponding to the child node from the historical process speech; obtaining the second candidate speech information output by the speech model; obtaining the similarity between the second candidate speech information and each of the candidate items; and determining the first candidate speech information from the at least two candidate items based on the similarity.

[0137] After that, we can take the latest customer reply (e.g., "Customer: I'm out of town, let's talk when I get back.") as the given context, and each candidate in candidate_text as the context, and combine them into three pairs of contexts. We can calculate the perplexity of the corresponding context based on the context (which can be used to evaluate the fluency and rationality of the text, the lower the value, the more natural the text). For each candidate, we set a text sequence X after word segmentation = (x0, x1, ..., x t ), the perplexity PPL(X) can be calculated as follows:

[0138]

[0139] where p θ (x i ∣x <i ) represents the sequence x based on i <i Get the i-th word x i The conditional probability of can be used to calculate the log-likelihood of the i-th token. i Indicates the i-th word in the word sequence, x <i represents all the participles before the i-th participle; p θ (x i ∣x <i ) means that at a given x i In the case of all previous participles, the i-th participle x i The conditional probability of .

[0140] Regarding the text sequence after word segmentation, that is, the text sequence after each pair of context word segmentation, for example: ["customer", ":", "I", "in", "out of town", "what", ","", "wait", "I", "go back", "talk about it later", ".", "agent", ":", "you", "what", "month", "back", "here", "ah", "?"].

[0141] Based on the above, this solution can calculate the perplexity corresponding to each candidate in candidate_text by giving the above context, for example: [{'text':'Agent: When do you plan to come back? ', 'PPL':'850'},{'text':'Agent: Which month will you come back here? ', 'PPL':'320'},{'text':'Agent: When will you come back? I will make an appointment for you in advance', 'PPL':'1270'}]; and sort them in descending order according to the perplexity, and obtain the sorting result candidate_text_rank = [{'text':'Agent: Which month will you come back here? ', 'PPL':'320'},{'text':'Agent: When do you plan to come back? ', 'PPL':'850'},{'text':'Agent: When will you come back? I will make an appointment for you in advance', 'PPL':'1270'}].

[0142] 8) Screening techniques:

[0143] For the ranking results obtained in step 7), the first candidate in candidate_text_rank (the smallest value in the perplexity ranking) is selected. If the perplexity of this candidate is less than or equal to a threshold (e.g., 500), output_speech is returned (i.e., the candidate, such as "Agent: What month will you be back?"). Otherwise, a fallback speech (corresponding to the preset speech information) can be returned, such as "Sorry, I didn't hear you clearly. Please say it again." This returns the agent's speech based on the screening results.

[0144] The contents of 7) and 8) above may correspond to the above-mentioned determination of target speech information based on the first candidate speech information, including: obtaining the perplexity value between the user text and the first candidate speech information in the conversation text; the perplexity value is used to determine the text parameters of the first candidate speech information, and the text parameters include: at least one of a fluency parameter and a rationality parameter; and determining the target speech information based on the perplexity value. The determination of the target speech information based on the perplexity value includes: obtaining the perplexity value with the minimum value as the target perplexity value; if the target perplexity value meets the output condition, using the first candidate speech information corresponding to the target perplexity value as the target speech information; if the target perplexity value does not meet the output condition, using the preset speech information as the target speech information.

[0145] 9) Return to the agent's words:

[0146] Just return the final speech obtained above (i.e. output_speech).

[0147] From the above, in this solution, the contents involving the use of large models include the above 1), 3) and 8). Among them, for large models, basic models can be used, and there is no need to additionally train large models, thereby avoiding the high training cost of large models; this solution can also support process control through business process limitations.

[0148] The solution provided in the embodiment of this application can be adopted Figure 3 The functional framework shown in the figure is implemented, and the modules are introduced as follows:

[0149] 1) Historical dialogue labeling module:

[0150] Functional Description: This module reads historical conversation data and, in conjunction with the business process system tag_tree (including the main process and sub-processes), uses large models (such as Qwen and GPT) to tag agent conversation texts one by one. The tagging results can be used to identify process nodes and recall conversation texts in subsequent conversations.

[0151] Input: historical conversation data, business process system tag_tree;

[0152] Output: The process node label of each dialogue text (including the main process node and sub-process node).

[0153] 2) Process node identification module:

[0154] Functional Description: This module receives real-time conversation text (QA pairs) and, in conjunction with the business process system, uses a large model to analyze the current process node of the conversation. It also provides preliminary agent scripts based on the conversation content.

[0155] Input: real-time conversation text (QA pair), business process system tag_tree;

[0156] Output: process nodes of the current conversation (including main process nodes and sub-process nodes) and preliminary agent scripts.

[0157] 3) Process identification module:

[0158] Function description: This module can be used to determine whether the current dialogue stage is the "conclusion" node of the business process, or whether the sub-process has ended; based on the judgment result, the next processing flow is determined.

[0159] Input: Output of the process node identification module (including main process nodes and sub-process nodes);

[0160] Output: the result of judging whether it is the end statement or not, the result of judging whether the sub-process is ended or not.

[0161] 4) Agent Script Recall Module:

[0162] Function description: When a sub-process is not completed, this module can be used to recall similar agent scripts from the process script library as candidates based on the current sub-process node, using similarity and confusion.

[0163] Input: current sub-process node, process script library;

[0164] Output: Similar agent script candidates.

[0165] 5) Screening module:

[0166] Function description: This module can be used to determine whether the perplexity of the first candidate in the recalled speech candidate set is less than the threshold. If so, the speech is returned; otherwise, the fallback speech is returned.

[0167] Input: candidate options for the speech;

[0168] Output: Filtered agent scripts or fallback scripts.

[0169] As described above, the solution provided by the embodiment of the present application has the following beneficial effects:

[0170] 1) The large model can automatically analyze conversation content and label it according to the business process system, greatly reducing labor costs.

[0171] 2) Based on process node identification, process script library and script templates, by recalling the scripts of the corresponding process nodes or generating new process scripts, it is more controllable than directly generating scripts using large models, avoiding problems such as random answers and repeated questions caused by large model hallucinations.

[0172] 3) When business processes change, only the corresponding business process system needs to be updated and adjusted without retraining the large model, which reduces the complexity and cost of system maintenance.

[0173] In summary, this plan involves the following contents:

[0174] a. Introduction and integration of big model technology: This solution introduces big model technology to achieve intelligent analysis of conversation content, and accurately tags and intelligently polishes conversations based on the business process system, which not only improves work efficiency but also enhances conversation quality.

[0175] b. Process Node Identification and Script Recall Mechanism: This solution integrates a process node identification mechanism, a process script library, and script template design. It combines the large model generation capability and perplexity ranking to screen candidate scripts. This allows for accurate recall of scripts matching the current process node during a conversation, or the generation of new process scripts when necessary. This mechanism ensures the continuity and controllability of the conversation, effectively avoiding the hallucination problem that may arise from large models. Specifically, it includes:

[0176] a. Process control method:

[0177] This solution utilizes a process control approach, primarily integrating big model technology with established business process systems. The big model automatically determines the process node at which a conversation is located and determines the agent's script based on the corresponding template script or script library. This process control approach ensures smooth and controllable conversations.

[0178] b. Mechanism for recalling and generating speech techniques:

[0179] The speech recall and generation mechanism constructed in this solution enables the system to accurately recall the speech that matches the current process node. At the same time, it can also generate new process speech based on the current conversation context (corresponding to the process identification involved above) and business process requirements (corresponding to the business process system involved above). This mechanism can effectively circumvent the hallucination problems that may be caused by large models, such as giving irrelevant answers or repeating questions, thereby achieving higher conversation coherence and accuracy. In addition, when the business process is adjusted or updated, this solution does not require tedious retraining of the large model, which can reduce the complexity and cost of system maintenance.

[0180] The present application also provides an information processing device, such as Figure 4 Shown, including:

[0181] A first determining module 41 is used to determine the node information of the business process node where the conversation is currently located based on the conversation text;

[0182] A first acquisition module 42 is configured to acquire, for a child node under the main process node where the conversation is currently located, first candidate speech information in the historical process speech when the node information indicates that the child node under the main process node has not ended;

[0183] The second determination module 43 is used to determine target speech information based on the first candidate speech information.

[0184] The information processing device provided in the embodiment of the present application determines the node information of the business process node where the conversation is currently located based on the conversation text; when the node information indicates that the child node under the main process node where the conversation is currently located has not ended, the first candidate speech information in the historical process speech is obtained for the child node; based on the first candidate speech information, the target speech information is determined; it can support the automatic acquisition of accurate target speech, thereby minimizing problems such as high implementation cost, poor experience and low accuracy caused by manual participation.

[0185] Among them, the obtaining of the first candidate speech information in the historical process speech includes: obtaining at least two candidate items corresponding to the sub-node from the historical process speech; obtaining the second candidate speech information output by the speech model; obtaining the similarity between the second candidate speech information and each of the candidate items; and determining the first candidate speech information from the at least two candidate items based on the similarity.

[0186] In an embodiment of the present application, determining the target speech information based on the first candidate speech information includes: obtaining the confusion value between the user text and the first candidate speech information in the conversation text; the confusion value is used to determine the text parameters of the first candidate speech information, and the text parameters include: at least one of a fluency parameter and a rationality parameter; determining the target speech information based on the confusion value.

[0187] Wherein, determining the target speech information based on the confusion value includes: obtaining the confusion value with the smallest numerical value as the target confusion value; when the target confusion value meets the output condition, using the first candidate speech information corresponding to the target confusion value as the target speech information; when the target confusion value does not meet the output condition, using the preset speech information as the target speech information.

[0188] In an embodiment of the present application, determining the node information of the business process node where the conversation is currently located based on the conversation text includes: using a recognition model to determine the node information of the business process node where the conversation is currently located based on the conversation text.

[0189] Furthermore, the information processing device also includes: a first labeling module, which is used to use a labeling model to label the historical speech text according to historical dialogue data and business process information to obtain the historical process speech.

[0190] Among them, the implementation embodiments of the above-mentioned information processing method are all applicable to the embodiments of the information processing device and can achieve the same technical effects.

[0191] An embodiment of the present application also provides an information processing device, including a memory, a processor, and a program stored in the memory and executable on the processor; when the processor executes the program, the above-mentioned information processing method is implemented.

[0192] Among them, the implementation embodiments of the above-mentioned information processing method are all applicable to the embodiments of the information processing device and can achieve the same technical effects.

[0193] An embodiment of the present application further provides a readable storage medium on which a program is stored, and when the program is executed by a processor, the steps in the above-mentioned information processing method are implemented.

[0194] Among them, the implementation embodiments of the above-mentioned information processing method are all applicable to the embodiments of the readable storage medium and can achieve the same technical effects.

[0195] An embodiment of the present application also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, the various processes of the method embodiment of the above-mentioned information processing method are implemented, and the same technical effect can be achieved. To avoid repetition, they will not be repeated here.

[0196] It should be noted that many functional components described in this specification are referred to as modules in order to more particularly emphasize the independence of their implementation methods.

[0197] In embodiments of the present application, modules can be implemented in software so that they can be executed by various types of processors. For example, an identified executable code module can include one or more physical or logical blocks of computer instructions, for example, which can be constructed as objects, processes, or functions. Nevertheless, the executable code of the identified module does not need to be physically located together, but can include different instructions stored in different locations, which, when logically combined together, constitute the module and achieve the specified purpose of the module.

[0198] In fact, executable code module can be a single instruction or many instructions, and can even be distributed on a plurality of different code segments, distributed in the middle of different programs, and distributed across a plurality of memory devices.Similarly, operating data can be identified in the module, and can be implemented and organized in the data structure of any appropriate type according to any appropriate form.Described operating data can be collected as a single data set, or can be distributed in different locations (including on different storage devices), and can only be present on a system or network as an electronic signal at least in part.

[0199] When a module can be implemented using software, given the current state of hardware technology, those skilled in the art can build corresponding hardware circuits to implement the corresponding functions of the module, regardless of cost. The hardware circuits may include conventional very large scale integration (VLSI) circuits or gate arrays, as well as existing semiconductors such as logic chips and transistors, or other discrete components. Modules may also be implemented using programmable hardware devices, such as field programmable gate arrays, programmable array logic, or programmable logic devices.

[0200] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary personnel in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. An information processing method, characterized in that: include: Determine the node information of the business process node where the conversation is currently located based on the conversation text; When the node information indicates that the child node under the main process node where the conversation is currently located has not ended, obtaining the first candidate speech information in the historical process speech for the child node; Determine target speech information based on the first candidate speech information.

2. The information processing method according to claim 1, wherein: The obtaining of the first candidate speech information in the historical process speech includes: Obtain at least two candidate items corresponding to the child node from the historical process dialogue; Obtain the second candidate speech information output by the speech model; Obtaining similarities between the second candidate speech information and each of the candidate items; Based on the similarity, the first candidate speech information is determined from the at least two candidate items.

3. The information processing method according to claim 1, wherein: The determining target speech information according to the first candidate speech information includes: Obtaining a perplexity value between the user text and the first candidate speech information in the conversation text; the perplexity value is used to determine text parameters of the first candidate speech information, the text parameters including at least one of a fluency parameter and a rationality parameter; Target speech information is determined based on the confusion value.

4. The information processing method according to claim 3, wherein: Determining target speech information according to the confusion value includes: Obtaining the minimum perplexity value as the target perplexity value; When the target perplexity value satisfies the output condition, the first candidate speech information corresponding to the target perplexity value is used as the target speech information; When the target perplexity value does not satisfy the output condition, the preset speech information is used as the target speech information.

5. The information processing method according to claim 1, wherein: Determining the node information of the business process node where the conversation is currently located based on the conversation text includes: Using the recognition model, the node information of the business process node where the conversation is currently located is determined based on the conversation text.

6. The information processing method according to claim 1, wherein: Also includes: Using the labeling model, the historical dialogue text is labeled according to the historical dialogue data and business process information to obtain the historical process dialogue.

7. An information processing device, characterized in that include: A first determination module is used to determine the node information of the business process node where the conversation is currently located based on the conversation text; A first acquisition module is configured to acquire, for a child node under a main process node where the conversation is currently located, first candidate speech information in historical process speech for the child node when the node information indicates that the child node has not ended; The second determination module is used to determine target speech information based on the first candidate speech information.

8. An information processing device comprising a memory, a processor, and a program stored in the memory and executable on the processor; When the processor executes the program, the information processing method according to any one of claims 1 to 6 is implemented.

9. A readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of the information processing method according to any one of claims 1 to 6 are realized.

10. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the information processing method according to any one of claims 1 to 6.

Images