WO2020119568A1 - 对话情感分析方法及装置、存储介质和处理器 - Google Patents

对话情感分析方法及装置、存储介质和处理器 Download PDF

Info

Publication number
WO2020119568A1
WO2020119568A1 PCT/CN2019/123215 CN2019123215W WO2020119568A1 WO 2020119568 A1 WO2020119568 A1 WO 2020119568A1 CN 2019123215 W CN2019123215 W CN 2019123215W WO 2020119568 A1 WO2020119568 A1 WO 2020119568A1
Authority
WO
WIPO (PCT)
Prior art keywords
analysis
message
user
emotional
intent
Prior art date
Application number
PCT/CN2019/123215
Other languages
English (en)
French (fr)
Inventor
林君
宋凯嵩
沈忱林
孙常龙
刘晓钟
Original Assignee
阿里巴巴集团控股有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 阿里巴巴集团控股有限公司 filed Critical 阿里巴巴集团控股有限公司
Publication of WO2020119568A1 publication Critical patent/WO2020119568A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35—Clustering; Classification

Definitions

  • the present invention relates to the field of computers, and in particular, to a dialogue sentiment analysis method and device, storage medium, and processor.
  • the communication process between consumers and customer service is an important reference data for merchants to judge the emotional tendencies of consumers; while in related technologies, the emotional tendency of the dialogue content between consumers and customer service requires the customer service to pass the current dialogue Content to understand and judge consumers' emotional tendencies, and the judgment of customer service is often subjective, which will lead to judgment errors.
  • Embodiments of the present invention provide a dialogue sentiment analysis method and device, a storage medium, and a processor, so as to at least solve the emotional tendency of the dialogue content between consumers and customer service in related technologies, and there is subjective presumption of customer service's own understanding and judgment technical problem.
  • a dialog sentiment analysis method including: acquiring dialog content between a first user and a second user, where the first user is a service provider and the second The user is a service object; perform a separate sentiment analysis on each message in the conversation content to obtain a first analysis result and perform an overall sentiment analysis on the conversation content based on the first analysis result of the sentiment analysis on each message to obtain a second Analysis results; at least according to the first analysis results and the second analysis results, respectively, determine the first user and the second user's emotional change trend in the dialogue content.
  • a dialogue sentiment analysis device including: an acquisition module for acquiring dialogue content between a first user and a second user, wherein the first user is a service provider, The second user is a service object; an analysis module is used to perform a separate sentiment analysis on each message in the conversation content to obtain a first analysis result, and based on the first analysis result of each message sentiment analysis on the conversation The overall sentiment analysis is performed on the content to obtain a second analysis result; a determination module is used to determine the content of the conversation between the first user and the second user respectively based on at least the first analysis result and the second analysis result The trend of emotional changes in
  • a storage medium includes a stored program, wherein, when the program is running, the device on which the storage medium is located is controlled to execute the above-mentioned dialog sentiment analysis method.
  • a processor for running a program, wherein the above-mentioned dialog sentiment analysis method is executed when the program is running.
  • each message in the conversation content of the first user of the service provider and the second user of the service object can be subjected to a separate sentiment analysis to obtain a first analysis result and the conversation content based on the first analysis result Perform an overall sentiment analysis to obtain a second analysis result.
  • Based on the first analysis result and the second analysis result determine the first user and second user's emotional change trend in the conversation content, and quickly determine the intentions, attitudes, and satisfactions of both parties. And so on, so that the two parties can adjust the subsequent dialogue content according to the emotional change trend, thereby solving the emotional tendency of the dialogue content between the consumer and the customer service in the related technology, and the customer service’s own understanding and judgment have subjective assumptions, which has improved communication. Success rate effect.
  • FIG. 1 shows a block diagram of a hardware structure of a computer terminal (or mobile device) for implementing a dialogue sentiment analysis method
  • FIG. 2 is a flowchart of a method for conversation sentiment analysis according to an embodiment of the present invention
  • FIG. 3 is a schematic structural diagram of a dialogue sentiment analysis device according to an embodiment of the present invention.
  • FIG. 4 is a schematic diagram of an optional structure of a dialogue sentiment analysis device according to an embodiment of the present invention.
  • an embodiment of a dialogue sentiment analysis method is also provided. It should be noted that the steps shown in the flowchart in the drawings can be executed in a computer system such as a set of computer-executable instructions, and Although the logical sequence is shown in the flowchart, in some cases, the steps shown or described may be performed in an order different from here.
  • FIG. 1 shows a hardware block diagram of a computer terminal (or mobile device) used to implement a conversation sentiment analysis method.
  • the computer terminal 10 may include one or more (shown as 102a, 102b, ..., 102n in FIG. 1) processor 102 (the processor 102 may include but is not limited to A processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission module for communication functions.
  • a processing device such as a microprocessor MCU or a programmable logic device FPGA
  • the computer terminal 10 can further include more or fewer components than those shown in FIG. 1, or have a configuration different from that shown in FIG.
  • the one or more processors 102 and/or other data processing circuits described above may generally be referred to herein as "data processing circuits.”
  • the data processing circuit may be fully or partially embodied as software, hardware, firmware, or any other combination.
  • the data processing circuit may be a single independent processing module, or may be wholly or partially integrated into any one of the other elements in the computer terminal 10 (or mobile device).
  • the data processing circuit serves as a kind of processor control (for example, selection of a variable resistance terminal path connected to an interface).
  • the memory 104 may be used to store software programs and modules of application software, such as the program instruction/data storage device corresponding to the () method in the embodiment of the present invention, and the processor 102 executes the software programs and modules stored in the memory 104 to execute Various functional applications and data processing, that is, to realize the vulnerability detection method of the above application program.
  • the memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory.
  • the memory 104 may further include memories remotely provided with respect to the processor 102, and these remote memories may be connected to the computer terminal 10 through a network. Examples of the above network include but are not limited to the Internet, intranet, local area network, mobile communication network, and combinations thereof.
  • the transmission device is used to receive or send data via a network.
  • the above specific example of the network may include a wireless network provided by a communication provider of the computer terminal 10.
  • the transmission device includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through the base station to communicate with the Internet.
  • the transmission device may be a radio frequency (RF) module, which is used to communicate with the Internet in a wireless manner.
  • RF radio frequency
  • the display may be, for example, a touch screen liquid crystal display (LCD), which may enable a user to interact with the user interface of the computer terminal 10 (or mobile device).
  • LCD liquid crystal display
  • FIG. 2 is a flowchart of a dialog sentiment analysis method according to an embodiment of the present invention. As shown in FIG. 2, the steps of the method include:
  • Step S202 Obtain the dialogue content between the first user and the second user, where the first user is the service provider and the second user is the service object;
  • the first user is a service provider
  • the second user is a service object
  • the first user may be a customer service of a merchant on the online shopping platform
  • the first The second user is an online shopping user who consults the merchant's goods. If the merchandise of the merchant is clothes, the content of the conversation between the first user and the second user may be: the material, size, discount, etc. of a certain piece of clothing consulted by the second user, and the first user Relevant content in response to questions asked by users.
  • the above-mentioned first user and second user are not limited to users of online shopping platforms, but also users of other consulting service platforms.
  • Step S204 Perform a separate sentiment analysis on each message in the conversation content to obtain a first analysis result, and perform an overall sentiment analysis on the conversation content based on the first analysis result of each message sentiment analysis to obtain a second analysis result;
  • the dialogue content is that the consumer asks: "Which size do you usually wear for pants?"; The customer service returns: "Please ask your height How many centimeters?” It can be seen that in this conversation, the consumer asked questions, and the customer service responded to the questions. There were no other polite or favorably tone words. Therefore, a separate sentiment analysis is performed on each message of this conversation, and the first analysis result obtained is neutral, that is, the sentiment degree of each message is neutral, so that the overall sentiment can also be obtained based on the first analysis result The degree is also neutral, that is, the change trend is neutral.
  • the customer's attitude is also It gradually improved, and finally returned the goods peacefully, and chose other products to replace. It can be seen that the first analysis result for the customer is from negative to neutral to the final positive, so that the second analysis result of the customer's overall sentiment analysis is positive, and then the emotional change trend in the entire dialogue process.
  • step S206 the emotional change trend of the first user and the second user in the dialogue content is determined according to at least the first analysis result and the second analysis result.
  • the emotional change trend may be neutral at the beginning.
  • the analysis of the message in the conversation content shows that the conversation content at this time is negative. That is to say, according to the content of the previous dialogue, the emotion of the second user changes from neutral to negative. After seeing this emotional change, the customer service concluded that if they continue to ignore it or have a bad attitude, the emotion of the second user will evolve in a more negative trend. Therefore, according to this emotional change trend, the customer service can respond to the message with a correct attitude or actively, and respond to the question patiently, so that the content of the dialogue between the two parties returns to a positive result.
  • each message in the conversation content of the first user of the service provider and the second user of the service object can be sentimentally analyzed to obtain a first analysis result and the dialogue based on the first analysis result
  • the overall sentiment analysis of the content results in a second analysis result.
  • the first user and the second user's emotional change trend in the conversation content can be judged, which can quickly determine the intentions, attitudes, and satisfactions of both parties.
  • the way to obtain the analysis result can be through To achieve as follows:
  • Step S204-11 identifying the intent corresponding to each message in the conversation content based on the text analysis model
  • the text analysis model can determine its intent through keywords or a whole sentence as a model.
  • the intent for the second user of the service object can include: greeting, regular question consultation, problem description, regular answer to customer service questions, requests Customer service operation, polite language, impatient question consultation, improper attitude response, etc.; and the first user corresponding to the intent for providing services can include: welcome message, answer customer service questions, guide customer service operations, ask customer questions, polite language , Concluding remarks, improper responses, etc.
  • Step S204-12 Perform sentiment analysis based on the identified intent corresponding to each message to obtain a first analysis result, and perform an overall sentiment analysis on the conversation content based on the first analysis result to obtain the second analysis result.
  • the dialogue content is intended to ask questions and answer customer service questions. For example, consumers ask: "Which trousers are usually worn, which code to choose? "Customer service back: "What is the height in centimeters?" It can be seen that in this conversation, the consumer asked questions, and the customer service responded to the questions. There were no other polite or friendly words. Therefore, sentiment analysis is performed on each message of this conversation, and the first analysis result of each message obtained is neutral, that is, the sentiment degree of each message is neutral, thus obtaining the overall sentiment analysis of the conversation content The second analysis result is also neutral. Of course, there are also emotions that are positive or negative.
  • the positive content is that the dialogue content is intended to consult and answer customer service questions.
  • the consumer asks: “Trouble sending me a delivery tomorrow morning", and the customer service returns "Yes, pro” .
  • the customer service responded positively to the consumer's question, and attached a "pro" with a respectful name, so that the consumer was more satisfied with the customer service answer.
  • the content of the dialogue is the request for customer service operation and the response to the operation conclusion. After the consumer sent a few messages without receiving a response: "Is it too lazy to read the information?" The customer service responded: "Yes.”
  • the conclusions about consumer dissatisfaction and poor customer service attitude are analyzed.
  • the first analysis result is obtained based on the sentiment analysis of each message, and then the overall sentiment analysis result of the conversation content can be obtained based on the first analysis result.
  • the first analysis result of sentiment analysis according to each message in front of the customer is: Emotional tendencies are all negative.
  • the subsequent customer's emotions are positive, then the customer's emotions of the entire dialogue are also positive, that is, the second analysis result is positive.
  • the customer's emotions are always negative and there is no change, then the customer's emotions in the entire dialogue are also negative, that is, the second analysis result is negative.
  • each message in the dialog content recognition based on the text analysis model corresponds to Before intent, the method steps of this embodiment may further include:
  • Step S208 Use the sample data to train the text analysis model and output a pre-judgment intent label, where the sample data includes one of the following: a message sample associated with the first user and a message sample associated with the second user;
  • the message sample associated with the first user and the message sample associated with the second user in the above sample data in a specific implementation manner, for example, the message sample associated with the first user "Pro, you report Your height and weight", “Dear, our products are 100% cotton", “Our store supports seven days of unreasonable return”, “Shipping on the day of payment before 3 pm on the same day”; associated with the second user
  • the above message samples are only examples, and can also be other message samples. The more samples, the better. Among them, it is better to have a relatively balanced proportion of different emotional levels.
  • the emotional levels are divided into: negative, neutral, and Positive; the samples of these three sentiment levels are all covered as much as possible, so as to ensure that the output pre-judgment intent label is reasonable, and then for subsequent sentiment analysis.
  • the division of the emotional level is only an example, and the emotional level can be further subdivided according to the actual situation.
  • negative can be divided into: very negative and general negative; positive can be divided into general positive and very negative positive. That is to say, the division of the emotion degree can be set and adjusted according to the actual emotion, and the above is only an example.
  • Step S210 comparing the pre-judgment intent label with the target intent label to obtain a comparison result
  • step S212 the model parameters used in the text analysis model are adjusted according to the comparison result, and the sample text data is used again to repeatedly train the adjusted text analysis model until the pre-judgment intent label matches the target intent label.
  • the model parameters can adjust the sample data, and then adjust the adjusted The sample data is repeatedly trained until the pre-judgment intent label matches the target intent label. That is to say, through step S210 and step S212, the intent label in the text analysis model can be matched with the target intent label to obtain the final desired text analysis model.
  • the manner of identifying the intent corresponding to each message in the conversation content based on the text analysis model involved in step S204-11 in this embodiment may include:
  • Steps S204-111 determining the source of each message in the conversation content
  • the source of the message refers to who sent the message in the dialogue content, that is, the first user (customer service) or the second user (consumer/customer). For example, in a conversation:
  • Steps S204-112 each message in the dialogue content is input to a text analysis model corresponding to the source of the message, and a pre-judgment intent label is output;
  • Steps S204-113 identifying the intent corresponding to each message in the conversation content according to the pre-judgment intent tag.
  • the pre-judgment intent tag of the message can be further output to determine the intent corresponding to each message.
  • the intention is to consult the question; Sending out “about when will it be sent?”, “don't do it as soon as possible!!” Enter the corresponding text analysis model, the intention you get is impatient.
  • an individual sentiment analysis and an overall sentiment analysis of the dialogue content are performed on the identified intent corresponding to each message involved in step S204-12 of this embodiment to obtain an analysis
  • Steps S204-121 determine the associated emotional tag from the emotional multi-tag classification relationship, wherein the emotional multi-tag classification relationship is used to maintain the relationship between the intent tag and the emotional tag corresponding to each message ,
  • the multi-label classification of emotions includes: at least one level of negative emotion tags, neutral emotion tags, and at least one level of positive emotion tags;
  • the second user intended for the service object in this embodiment may include: hello, general question consultation, problem description, general answer to customer service questions, request customer service operations, polite language, no Patience consultation, improper attitude response, etc.; and the first users corresponding to the intention to provide services can include: welcome message, answer customer service questions, guide customer service operations, ask customer questions, polite words, concluding remarks, improper attitude response, etc. Wait.
  • the associated emotional tags are determined according to the emotional multi-label classification relationship
  • Customer Service The warehouse will step up arrangements for processing. (Positive response to customers, positive emotional tags)
  • negative emotion tags may include general negative emotion tags and very negative emotion tags
  • positive emotion tags may include: general positive emotion tags and very positive emotion tags.
  • sentiment tags can also be a more detailed division of sentiment tags. The above is only an example.
  • Steps S204-122 Determine the first analysis result and the second analysis result according to the emotion level corresponding to the emotion tag.
  • the emotion level of each message can be determined according to the emotion tag of each message in the conversation content, and the emotion tag based on each message can determine the customer as a whole (second The overall emotions of the customer) and the customer service (first user).
  • the customer service reflects the impatient and anxious negative emotions. Therefore, the overall emotion of the customer is negative, and the customer service has always responded positively. Therefore, the overall emotion of the customer service is positive.
  • the following dialogue is a dialogue in which the overall emotions of the customer and the customer service are negative.
  • step S206 of this embodiment to separately determine the emotional change trend of the first user and the second user in the conversation content according to the analysis results may be as follows achieve:
  • Step S206-11 the emotion degree corresponding to each message in the conversation content is counted by the first analysis result to obtain a first statistical result, and the emotion change of the conversation content according to time is counted by the second analysis result, Get the second statistical result;
  • the sentiment level of each message of the entire dialogue content is counted, that is, each message has a corresponding sentiment tag to obtain the first statistical result.
  • the time node generated by each emotion tag is marked, that is, at which time period of the dialogue content each emotion tag needs to be generated, and then the first analysis result and the dialogue content can be based on The emotional change of the time progress is counted to obtain the second statistical result.
  • Step S206-12 at least according to the first statistical result and the second statistical result, respectively, determine the emotional change trend of the first user and the second user in the dialogue content.
  • the emotional change trend of the first user and the second user is determined according to the first statistical result and the second statistical result.
  • the current overall emotion of the second user is negative.
  • the overall emotion of the second user may still be negative. Therefore, the first user may change the response content according to the trend of the overall emotion of the second user.
  • the customer's initial tendency is negative, and after the customer service treatment, it is positive, so the emotion of the customer in the entire dialogue is also positive.
  • the method according to the above embodiments can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware, but in many cases the former is Better implementation.
  • the technical solutions of the present invention can be embodied in the form of software products in essence or part of contributions to the existing technology, and the computer software products are stored in a storage medium (such as ROM/RAM, magnetic disk,
  • the CD-ROM includes several instructions to enable a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the embodiments of the present invention.
  • FIG. 3 is a schematic structural diagram of a dialog sentiment analysis device according to an embodiment of the present invention. As shown in FIG. 3, the device include:
  • the obtaining module 32 is used to obtain the dialogue content between the first user and the second user, wherein the first user is a service provider and the second user is a service object;
  • the first user is a service provider
  • the second user is a service object
  • the first user may be a customer service of a merchant on the online shopping platform
  • the first The second user is an online shopping user who consults the merchant's goods. If the merchandise of the merchant is clothes, the content of the conversation between the first user and the second user may be: the material, size, discount, etc. of a certain piece of clothing consulted by the second user, and the first user Relevant content in response to questions asked by users.
  • the above-mentioned first user and second user are not limited to users of online shopping platforms, but also users of other consulting service platforms.
  • the analysis module 34 is coupled to the acquisition module 32, and is used to perform a separate sentiment analysis on each message in the dialog content to obtain a first analysis result, and perform an overall sentiment analysis on the dialog content based on the first analysis result of the sentiment analysis on each message. To get the second analysis result;
  • the dialogue content is, the consumer asks: "Which size do you usually wear for pants?"; The customer service returns: "Please ask your height How many centimeters?” It can be seen that in this conversation, the consumer asked questions, and the customer service responded to the questions. There were no other polite or favorably tone words. Therefore, a separate sentiment analysis is performed on each message of this conversation, and the first analysis result obtained is neutral, that is, the sentiment degree of each message is neutral. Of course, there are also emotions that are positive or negative, for example, the positive is that consumers ask: "Trouble sending me a courier tomorrow morning.” The customer service responded "Yes, pro".
  • a second analysis result that characterizes the overall sentiment tendency can be obtained based on the first analysis result of the sentiment analysis of each specific message, for example,
  • the customer's initial tone and conversation content are not friendly, and the words are relatively sharp, but the customer service conversation is very polite.
  • the customer's attitude is also It gradually improved, and finally returned the goods peacefully, and chose other products to replace. It can be seen that the first analysis result for the customer is from negative to neutral to the final positive, so that the second analysis result of the customer's overall sentiment analysis is positive, and then the emotional change trend in the entire dialogue process.
  • the determination module 36 is coupled to the analysis module 34, and is used to determine the emotional change trend of the first user and the second user in the dialogue content according to at least the first analysis result and the second analysis result.
  • the emotional change trend may be neutral at the beginning.
  • the analysis of the message in the conversation content shows that the conversation content at this time is negative. That is to say, according to the content of the previous dialogue, the emotion of the second user changes from neutral to negative. After seeing this emotional change, the customer service concluded that if they continue to ignore it or have a bad attitude, the emotion of the second user will evolve in a more negative trend. Therefore, according to this emotional change trend, the customer service can respond to the message with a correct attitude or actively, and respond to the question patiently, so that the content of the dialogue between the two parties returns to a positive result.
  • the above acquisition module 32, analysis module 34, and determination module 36 correspond to steps S202 to S206 in Embodiment 1, and the two modules and the corresponding steps implement the same examples and application scenarios, but not It is limited to the content disclosed in the above first embodiment. It should be noted that the above-mentioned modules can be run in the computer terminal 10 provided in the first embodiment as a part of the device.
  • the analysis module 34 in this embodiment may include: an identification unit, coupled to the analysis unit, for identifying the intent corresponding to each message in the conversation content based on the text analysis model;
  • the recognition unit can determine its intent based on the keywords of the text analysis model or the whole sentence as a model, and the intent can include: greetings, routine question consultation, question description, and routine customer service questions for the second user of the service object , Request customer service operations, polite language, impatient question consultation, improper attitude response, etc.; and the first user corresponding to the intention for providing services can include: welcome message, answer customer service questions, guide customer service operations, ask customer questions, Polite words, concluding remarks, improper responses, etc. It should be noted that the above intentions are only examples, and other intentions can also be included. The above examples do not constitute a limitation on the present application.
  • the analysis unit is configured to perform sentiment analysis based on the identified intent corresponding to each message to obtain a first analysis result, and perform an overall sentiment analysis on the dialogue content based on the first analysis result to obtain the second analysis result.
  • the dialogue content is intended to ask questions and answer customer service questions. For example, consumers ask: "Which trousers are usually worn, which code to choose? "Customer service back: "What is the height in centimeters?" It can be seen that in this conversation, the consumer asked questions, and the customer service responded to the questions. There were no other polite or friendly words. Therefore, the analysis unit performs sentiment analysis on each message of this conversation, and the first analysis result of each message obtained is neutral, that is, the sentiment degree of each message is neutral, thereby obtaining the overall content of the conversation The second analysis result of sentiment analysis is also neutral. Of course, there are also emotions that are positive or negative.
  • the positive content is that the dialogue content is intended to consult and answer customer service questions.
  • the consumer asks: "Trouble sending me a delivery tomorrow morning", and the customer service returns "Yes, pro” .
  • the customer service responded positively to the customer's question, and attached a "pro" with a respectful name, so that the consumer was satisfied with the customer service answer.
  • the content of the dialogue is the request for customer service operation and the response to the operation conclusion. After the consumer sent a few messages without receiving a response: "Is it too lazy to read the information?" The customer service responded: "Yes.” According to the intent corresponding to the message, the conclusion of consumer dissatisfaction and poor customer service attitude is analyzed.
  • the first analysis result is obtained based on the sentiment analysis of each message, and then the overall sentiment analysis result of the conversation content can be obtained based on the first analysis result.
  • the first analysis result of sentiment analysis according to each message in front of the customer is: Emotional tendencies are all negative.
  • the subsequent customer's emotions are positive, then the customer's emotions of the entire dialogue are also positive, that is, the second analysis result is positive.
  • the customer's emotions are always negative and there is no change, then the customer's emotions are also negative throughout the dialogue, that is, the second analysis result is negative.
  • FIG. 4 is a schematic diagram of an optional structure of a dialogue sentiment analysis device according to an embodiment of the present invention. As shown in FIG. 4, the device further includes:
  • the first training module 42 is coupled to the comparison module 44 and used to train the text analysis model using sample data before outputting the intent label based on the text analysis model before identifying the intent corresponding to each message in the dialogue content based on the text analysis model.
  • the sample data includes one of the following: a message sample associated with the first user and a message sample associated with the second user;
  • the message sample associated with the first user and the message sample associated with the second user in the above sample data in a specific implementation manner, for example, the message sample associated with the first user "Pro, you report Your height and weight”, “Pro, our products are 100% cotton", “Our store supports seven days of unreasonable return”, “Shipping on the day of payment before 3 pm on the same day”; associated with the second user
  • the above message samples are only examples, and can also be other message samples. The more samples, the better. Among them, it is better to have a relatively balanced proportion of different emotional levels.
  • the emotional levels are divided into: negative, neutral, and Positive; the samples of these three sentiment levels are all covered as much as possible, so as to ensure that the output pre-judgment intent label is reasonable, and then for subsequent sentiment analysis.
  • the division of the emotional level is only an example, and the emotional level can be further subdivided according to the actual situation.
  • negative can be divided into: very negative and general negative; positive can be divided into general positive and very negative positive. That is to say, the division of the emotion degree can be set and adjusted according to the actual emotion, and the above is only an example.
  • the comparison module 44 is coupled to the second training module 46, and is used to compare the pre-determined intent label with the target intent label to obtain a comparison result;
  • the second training module 46 is coupled to the analysis module 34 for adjusting the model parameters used by the text analysis model according to the comparison result, and re-training the adjusted text analysis model using sample data until the pre-judgment intent label Match the target intent label.
  • the identification unit involved in this embodiment includes:
  • the first determining subunit is used to determine the source of each message in the conversation content
  • the source of the message refers to who sent the message in the content of the conversation, that is, whether it was sent by the first user (customer service) or by the second user (consumer/customer), for example, in a conversation:
  • the first determination sub-unit can determine in the above dialogue "When will the goods be delivered”, “When will the delivery be delivered?", “When should it be delivered?”, “Don't do it as soon as possible!!!”, “Well, please trouble as soon as possible” It was sent by the customer.
  • the first sub-unit confirms that "existing, pro”, “pro, courier default Shentong, Yuantong, postal, there is no other courier temporarily, nor accept designated courier, please understand that”, “pro, here will I will arrange it for you as soon as possible”, “Pro, the specific time depends on the arrangement on the side of the warehouse” "The warehouse will step up the arrangement", “Okay", these are all responded or sent by the customer service .
  • the input subunit is coupled to the first determining subunit, and is used to input each message in the conversation content to a text analysis model corresponding to the source of the message, and output a pre-judgment intent label;
  • the recognition subunit is coupled to the input subunit, and is used for recognizing the intent corresponding to each message in the conversation content according to the pre-determined intent label.
  • the pre-judged intent label of the message can be further output by the input subunit, and then the subunit determines the intent corresponding to each message.
  • the intent is to ask questions; Time to send?”, "Don't do it as soon as possible!!” Enter the corresponding text analysis model, the intention you get is impatient.
  • the analysis unit in this embodiment includes:
  • the first determining subunit is used to determine the associated sentiment tags from the sentiment multi-tag classification relationship according to the identified intent corresponding to each message, wherein the sentiment multi-tag classification relationship is used to maintain the intent tags and sentiments corresponding to each message
  • the sentiment multi-tag classification includes: at least one level of negative sentiment tags, neutral sentiment tags, and at least one level of positive sentiment tags;
  • the second user intended for the service object in this embodiment may include: hello, general question consultation, problem description, general answer to customer service questions, request customer service operations, polite language, no Patience consultation, improper attitude response, etc.; and the first users corresponding to the intention to provide services can include: welcome message, answer customer service questions, guide customer service operations, ask customer questions, polite words, concluding remarks, improper attitude response, etc. Wait.
  • the first determining subunit determines the associated sentiment tags according to the sentiment multi-tag classification relationship
  • Customer Service The warehouse will step up arrangements for processing. (Positive response to customers, positive emotional tags)
  • negative emotion tags may include general negative emotion tags and very negative emotion tags
  • positive emotion tags may include: general positive emotion tags and very positive emotion tags.
  • sentiment tags can also be a more detailed division of sentiment tags. The above is only an example.
  • the third determining subunit is used to determine the first analysis result and the second analysis result according to the emotion level corresponding to the emotion tag.
  • the third determining subunit can determine the sentiment level of each message according to the sentiment label of each message in the dialog content, and the sentiment label based on each message can determine the customer as a whole ( The overall emotions of the second user) and the customer service (first user).
  • the customer service reflects the impatient and anxious negative emotions, so the overall emotion of the customer is negative, and the customer service has always responded positively. , So the customer service sentiment is positive.
  • the following dialogue is a dialogue in which the overall emotions of the customer and the customer service are negative.
  • the determination module 36 involved in this embodiment includes:
  • the statistical unit is configured to perform statistics on the sentiment degree corresponding to each message in the dialog content through the first analysis result to obtain a first statistical result, and on the second analysis result to count the emotional changes of the dialog content according to time, Get the second statistical result;
  • the statistical unit may be to count the sentiment level of each message of the entire dialogue content, that is, each message has a corresponding sentiment tag to obtain the first statistical result. Then, based on the entire conversation content, the time node generated by each emotion tag is marked, that is, at which time period of the conversation content each emotion tag needs to be generated, and then based on the first analysis result and the conversation content The emotional change of the time progress is counted to obtain the second statistical result.
  • the determining unit is configured to determine the emotional change trend of the first user and the second user in the dialogue content at least according to the first statistical result and the second statistical result.
  • the emotional change trend of the first user and the second user is determined according to the first statistical result and the second statistical result.
  • the current overall emotion of the second user is negative.
  • the overall emotion of the second user may still be negative. Therefore, the first user can change the response content according to the trend of the overall emotion of the second user.
  • the customer's initial tendency is negative, and after the customer service treatment, it is positive, so the emotion of the customer in the entire dialogue is also positive.
  • An embodiment of the present invention may provide a processor, which may be used in a computer terminal, and the computer terminal may be any computer terminal device in a computer terminal group.
  • the above computer terminal may also be replaced with a terminal device such as a mobile terminal.
  • the foregoing computer terminal may be located in at least one network device among multiple network devices of the computer network.
  • FIG. 1 is a structural block diagram of a computer terminal according to an embodiment of the present invention.
  • the computer terminal A may include one or more processors, memory, and interfaces.
  • the memory can be used to store software programs and modules, such as program instructions/modules corresponding to the security vulnerability detection method and device in the embodiments of the present invention, and the processor executes various functions by running the software programs and modules stored in the memory Application and data processing, that is, to realize the detection method of the above system vulnerability attack.
  • the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory.
  • the memory may further include memories remotely provided with respect to the processor, and these remote memories may be connected to the terminal A through a network. Examples of the above network include but are not limited to the Internet, intranet, local area network, mobile communication network, and combinations thereof.
  • the processor can call the information and application stored in the memory through the transmission device to perform the following steps:
  • the processor may also execute the program code of the following steps: identify the intent corresponding to each message in the dialogue content based on the text analysis model; perform sentiment analysis based on the identified intent corresponding to each message to obtain the first 1. An analysis result, and an overall sentiment analysis of the dialogue content based on the first analysis result to obtain a second analysis result.
  • the above processor may also execute the program code of the following steps: before identifying the intent corresponding to each message in the dialog content based on the text analysis model, use sample data to train the text analysis model and output a pre-judgment intent label,
  • the sample data includes one of the following: a message sample associated with the first user and a message sample associated with the second user; comparing the pre-judged intent label with the target intent label to obtain a comparison result; adjusting according to the comparison result
  • the model parameters used by the text analysis model, and the sample data are used again to repeatedly train the adjusted text analysis model until the pre-judgment intent label matches the target intent label.
  • the above processor may also execute the program code of the following steps: determine the source of each message in the dialog content; input each message in the dialog content into the text analysis model corresponding to the source of the message, and output the pre-judgment Intent tags; identify the intent corresponding to each message in the conversation according to the pre-judged intent tags.
  • the processor may also execute the program code of the following steps: determine the associated emotion tags from the emotional multi-label classification relationship according to the identified intent corresponding to each message, wherein the emotional multi-label classification relationship is used to maintain each The mapping relationship between the intent tag and the emotional tag corresponding to the message.
  • the emotional multi-label classification includes: at least one level of negative emotional tags, neutral emotional tags, and at least one level of positive emotional tags; determined according to the degree of emotion corresponding to the emotional tag The first analysis result and the second analysis result.
  • the above processor may also execute the program code of the following steps: statistic the sentiment level corresponding to each message in the dialog content through the first analysis result to obtain a first statistical result, and use the second analysis result to analyze the dialog content Perform statistics according to the emotional change of the time schedule to obtain the second statistical result; at least determine the emotional change trend of the first user and the second user in the conversation content according to the first statistical result and the second statistical result respectively.
  • a solution for dialogue sentiment analysis is provided.
  • the emotional change trend of the first user and the second user in the conversation content can be determined, and the intention, attitude, satisfaction, etc. of both parties can be quickly determined, so that the two parties can
  • the emotional change trend adjusts the follow-up dialogue content, and thus solves the emotional tendency of the dialogue content between consumers and customer service in related technologies.
  • the customer service's own understanding and judgment have subjective and presumed technical problems, which has achieved the effect of improving the success rate of communication.
  • FIG. 10 is only an illustration, and the computer terminal may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, an applause computer, and a mobile Internet device (Mobile Internet Devices, MID ), PAD and other terminal equipment.
  • FIG. 10 does not limit the structure of the above electronic device.
  • the computer terminal 10 may also include more or fewer components than those shown in FIG. 10 (such as a network interface, a display device, etc.), or have a configuration different from that shown in FIG. 10.
  • the program may be stored in a computer-readable storage medium, and the storage medium may Including: flash disk, read-only memory (Read-Only Memory, ROM), random access device (Random Access Memory, RAM), magnetic disk or optical disk, etc.
  • Embodiments of the present invention also provide a storage medium.
  • the above storage medium may be used to store the program code executed by the method for dialog sentiment analysis provided in the above first embodiment.
  • the above-mentioned storage medium may be located in any computer terminal in the computer terminal group in the computer network, or in any mobile terminal in the mobile terminal group.
  • the storage medium is set to store program code for performing the following steps:
  • the storage medium may also execute the program code of the following steps: identify the intent corresponding to each message in the conversation content based on the text analysis model; perform sentiment analysis based on the identified intent corresponding to each message to obtain the first analysis As a result, and performing an overall sentiment analysis on the dialogue content based on the first analysis result, the second analysis result is obtained.
  • the above storage medium may also execute the program code of the following steps: before identifying the intent corresponding to each message in the dialog content based on the text analysis model, use sample data to train the text analysis model and output a pre-judgment intent label,
  • the sample data includes one of the following: a message sample associated with the first user and a message sample associated with the second user; comparing the pre-judged intent label with the target intent label to obtain a comparison result; adjusting according to the comparison result
  • the model parameters used by the text analysis model, and the sample data are used again to repeatedly train the adjusted text analysis model until the pre-judgment intent label matches the target intent label.
  • the above storage medium may also execute the program code of the following steps: determine the message source of each message in the conversation content; input each message in the conversation content into the text analysis model corresponding to the message source, and output the pre-judgment Intent tags; identify the intent corresponding to each message in the conversation according to the pre-judged intent tags.
  • the above storage medium may also execute the program code of the following steps: according to the identified intent corresponding to each message, determine the associated emotional tags from the emotional multi-tag classification relationship, where the emotional multi-tag classification relationship is used to maintain each The mapping relationship between the intent tag and the emotional tag corresponding to the message.
  • the emotional multi-label classification includes: at least one level of negative emotional tags, neutral emotional tags, and at least one level of positive emotional tags; determined according to the degree of emotion corresponding to the emotional tag The first analysis result and the second analysis result.
  • the above storage medium may also execute the program code of the following steps: statistic the sentiment level corresponding to each message in the conversation content through the first analysis result to obtain a first statistical result, and use the second analysis result to analyze the The content of the conversation is counted according to the emotional change of the time progress to obtain the second statistical result; at least the first and second statistical results are used to determine the emotional change trend of the first user and the second user in the conversation content.
  • the disclosed technical content may be implemented in other ways.
  • the device embodiments described above are only schematic.
  • the division of the unit is only a logical function division.
  • there may be another division manner for example, multiple units or components may be combined or may Integration into another system, or some features can be ignored, or not implemented.
  • the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, units or modules, and may be in electrical or other forms.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
  • the above integrated unit may be implemented in the form of hardware or software functional unit.
  • the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a computer-readable storage medium.
  • the technical solution of the present invention essentially or part of the contribution to the existing technology or all or part of the technical solution can be embodied in the form of a software product, the computer software product is stored in a storage medium , Including several instructions to enable a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention.
  • the foregoing storage media include: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program code .

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Machine Translation (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种对话情感分析方法及装置、存储介质和处理器。其中，该方法包括：获取第一用户与第二用户之间的对话内容（S202），其中，第一用户为服务提供者，第二用户为服务对象；对对话内容中的每条消息进行单独情感分析，得到第一分析结果，以及基于每条消息情感分析的第一分析结果对对话内容进行整体情感分析，得到第二分析结果（S204）；至少根据第一分析结果和第二分析结果分别确定第一用户与第二用户在对话内容中的情感变化趋势（S206）。该方法解决了相关技术中对于消费者与客服的对话内容的情感倾向，客服自己的理解和判断存在主观臆断的技术问题。

Description

对话情感分析方法及装置、存储介质和处理器
本申请要求2018年12月11日递交的申请号为201811512090.7、发明名称为“对话情感分析方法及装置、存储介质和处理器”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及计算机领域，具体而言，涉及一种对话情感分析方法及装置、存储介质和处理器。
背景技术
在日常的消费过程中，消费者与客服的沟通过程是商家判断消费者的情感倾向的重要参考数据；而相关技术中，对于消费者与客服的对话内容的情感倾向，需要客服通过当前的对话内容自己去理解和判断消费者的情感倾向，而在客服的判断往往存在主观臆断，进而会导致判断失误。
针对相关技术中的上述问题，目前尚未提出有效的解决方案。
发明内容
本发明实施例提供了一种对话情感分析方法及装置、存储介质和处理器，以至少解决了相关技术中对于消费者与客服的对话内容的情感倾向，客服自己的理解和判断存在主观臆断的技术问题。
根据本发明实施例的一个方面，提供了一种对话情感分析方法，包括：获取第一用户与第二用户之间的对话内容，其中，所述第一用户为服务提供者，所述第二用户为服务对象；对所述对话内容中的每条消息进行单独情感分析，得到第一分析结果以及基于每条消息情感分析的第一分析结果对所述对话内容进行整体情感分析，得到第二分析结果；至少根据所述第一分析结果和所述第二分析结果分别确定所述第一用户与所述第二用户在所述对话内容中的情感变化趋势。
根据本发明的另一个方面，提供了一种对话情感分析装置，包括：获取模块，用于获取第一用户与第二用户之间的对话内容，其中，所述第一用户为服务提供者，所述第二用户为服务对象；分析模块，用于对所述对话内容中的每条消息进行单独情感分析得到第一分析结果，以及基于每条消息情感分析的第一分析结果对所述对话内容进行整体 情感分析，得到第二分析结果；确定模块，用于至少根据所述第一分析结果和所述第二分析结果分别确定所述第一用户与所述第二用户在所述对话内容中的情感变化趋势。
根据本发明实施例的另一个方面，提供了一种存储介质，所述存储介质包括存储的程序，其中，在所述程序运行时控制所述存储介质所在设备执行上述的对话情感分析方法。
根据本发明实施例的另一个方面，提供了一种处理器，所述处理器用于运行程序，其中，所述程序运行时上述的对话情感分析方法。
在本发明实施例中，能够对服务提供者的第一用户与服务对象的第二用户的对话内容中的每一条消息进行单独情感分析得到第一分析结果以及基于该第一分析结果对对话内容进行整体情感分析得到第二分析结果，根据该第一分析结果和第二分析结果来判断第一用户和第二用户在对话内容中的情感变化趋势，能够快速确定双方的意向、态度、满意度等等，以使得双方根据该情感变化趋势调整后续对话内容，从而解决了相关技术中对于消费者与客服的对话内容的情感倾向，客服自己的理解和判断存在主观臆断的问题，达到了提高沟通成功率的效果。
附图说明
此处所说明的附图用来提供对本发明的进一步理解，构成本申请的一部分，本发明的示意性实施例及其说明用于解释本发明，并不构成对本发明的不当限定。在附图中：
图1示出了一种用于实现对话情感分析方法的计算机终端(或移动设备)的硬件结构框图；
图2是根据本发明实施例的对话情感分析方法的流程图；
图3是根据本发明实施例的对话情感分析装置的结构示意图；
图4是根据本发明实施例的对话情感分析装置的可选结构示意图。
具体实施方式
为了使本技术领域的人员更好地理解本发明方案，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分的实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本发明保护的范围。
需要说明的是，本发明的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本发明的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
实施例1
根据本发明实施例，还提供了一种对话情感分析方法的实施例，需要说明的是，在附图的流程图示出的步骤可以在诸如一组计算机可执行指令的计算机系统中执行，并且，虽然在流程图中示出了逻辑顺序，但是在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤。
本申请实施例一所提供的方法实施例可以在移动终端、计算机终端或者类似的运算装置中执行。图1示出了一种用于实现对话情感分析方法的计算机终端(或移动设备)的硬件结构框图。如图1所示，计算机终端10(或移动设备10)可以包括一个或多个(图1中采用102a、102b，……，102n来示出)处理器102(处理器102可以包括但不限于微处理器MCU或可编程逻辑器件FPGA等的处理装置)、用于存储数据的存储器104、以及用于通信功能的传输模块。除此以外，还可以包括：显示器、输入/输出接口(I/O接口)、通用串行总线(USB)端口(可以作为I/O接口的端口中的一个端口被包括)、网络接口、电源和/或相机。本领域普通技术人员可以理解，图1所示的结构仅为示意，其并不对上述电子装置的结构造成限定。例如，计算机终端10还可包括比图1中所示更多或者更少的组件，或者具有与图1所示不同的配置。
应当注意到的是上述一个或多个处理器102和/或其他数据处理电路在本文中通常可以被称为“数据处理电路”。该数据处理电路可以全部或部分的体现为软件、硬件、固件或其他任意组合。此外，数据处理电路可为单个独立的处理模块，或全部或部分的结合到计算机终端10(或移动设备)中的其他元件中的任意一个内。如本申请实施例中所涉及到的，该数据处理电路作为一种处理器控制(例如与接口连接的可变电阻终端路径的选择)。
存储器104可用于存储应用软件的软件程序以及模块，如本发明实施例中的()方法对应的程序指令/数据存储装置，处理器102通过运行存储在存储器104内的软件程序 以及模块，从而执行各种功能应用以及数据处理，即实现上述的应用程序的漏洞检测方法。存储器104可包括高速随机存储器，还可包括非易失性存储器，如一个或者多个磁性存储装置、闪存、或者其他非易失性固态存储器。在一些实例中，存储器104可进一步包括相对于处理器102远程设置的存储器，这些远程存储器可以通过网络连接至计算机终端10。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
传输装置用于经由一个网络接收或者发送数据。上述的网络具体实例可包括计算机终端10的通信供应商提供的无线网络。在一个实例中，传输装置包括一个网络适配器(Network Interface Controller，NIC)，其可通过基站与其他网络设备相连从而可与互联网进行通讯。在一个实例中，传输装置可以为射频(Radio Frequency，RF)模块，其用于通过无线方式与互联网进行通讯。
显示器可以例如触摸屏式的液晶显示器(LCD)，该液晶显示器可使得用户能够与计算机终端10(或移动设备)的用户界面进行交互。
在上述运行环境下，本申请提供了如图2所示的对话情感分析方法。图2是根据本发明实施例的对话情感分析方法的流程图，如图2所示，该方法的步骤包括：
步骤S202，获取第一用户与第二用户之间的对话内容，其中，第一用户为服务提供者，第二用户为服务对象；
需要说明的是，由于本实施例中的第一用户为服务提供者，第二用户为服务对象；因此，在具体应用场景中，该第一用户可以是网购平台某一商家的客服，而第二用户则为咨询该商家商品的网购用户。如果该商家的商品是衣服的话，则该第一用户与第二用户之间的对话内容可以是：第二用户咨询的某一件衣服的材质、大小、折扣等等，以及第一用户根据第二用户咨询的问题所回复的相关内容。当然，上述第一用户和第二用户并不局限于是网购平台的用户，也可以是其他咨询服务类平台的用户。当然，上述网购的具体应用场景仅仅是举例说明，也可以是其他能够场景，例如，电信服务业中客服与机主之间的对话内容，也可以是咨询服务业中的客服与顾客之间的对话内容，或者是其他场景中的对话内容等等。只要是存在两者之间的对话内容，则可以应用本申请进行情感倾向的分析。
步骤S204，对对话内容中的每条消息进行单独情感分析得到第一分析结果，以及基于每条消息情感分析的第一分析结果对对话内容进行整体情感分析，得到第二分析结果；
其中，以网购平台中消费者作为第二用户，商家客服作为第一用户为例，其对话内 容为，消费者问：“平时穿34的裤子，选哪个码子？”；客服回：“请问身高是多少厘米？”可见，这段对话内容中消费者问了问题，客服针对问题进行了回应，没有其他客套或增加好感的语气词。因此，对这段对话的每条消息进行单独情感分析，得到的第一分析结果为中性，即每条消息的情感程度均为中性，从而也可以基于该第一分析结果得到整体的情感程度也是中性，即变化趋势一致是中性。当然，也有情感为正面的或者是负面的，例如正面的是，消费者问：“麻烦明早帮我发快递”。客服回“可以的，亲”。负面的是，消费者在连发了几条信息都没有收到回应后：“是不是连信息都懒得看”，客服回应：“是的”。而对于一段对话内容有正面，负面、中性的情感程度的情况下，可以根据具体每条消息的情感分析得到的第一分析结果得到用于表征整体的情感倾向的第二分析结果，例如，在网购的退货对话内容中，顾客一开始的语气与谈话内容都是不友善的，用词也比较尖锐，但是客服的会话一致是非常客气的，在客服的友好的态度下，顾客的态度也慢慢变好了，最终和平的退了货，并选择了其他商品来代替。可见，对该顾客第一分析结果为情感由负面到中性直到最后的正面，这样就得到顾客的整体情感分析的第二分析结果为正面，进而得到整个对话过程中的情感变化趋势。
步骤S206，至少根据第一分析结果和第二分析结果分别确定第一用户与第二用户在对话内容中的情感变化趋势。
其中，该情感变化趋势可以是一开始是中性的，后续沟通过程中可能由于客服回复不及时或者客服态度不好，则对对话内容的中的消息分析得到此时的对话内容为负面。也就是说，根据前面对话内容，第二用户的情感由中性变为负面。客服看到这个情感变化之后，得到的结论是如果继续不理会或态度不好则第二用户的情感将会以更加负面的趋势演化。因此，根据该情感变化趋势，客服可以以端正态度或积极回复消息，并对问题进行耐心解释的方式进行回复，从而使得双方的对话内容回到了正面结果。
通过上述步骤S202至步骤S206可知，能够对服务提供者的第一用户与服务对象的第二用户的对话内容中的每一条消息进行情感分析得到第一分析结果以及基于该第一分析结果对对话内容进行整体情感分析得到第二分析结果，根据该第一分析结果和第二分析结果来判断第一用户和第二用户在对话内容中的情感变化趋势，能够快速确定双方的意向、态度、满意度等等，以使得双方根据该情感变化趋势调整后续对话内容，从而解决了相关技术中对于消费者与客服的对话内容的情感倾向，客服自己的理解和判断存在主观臆断的问题，达到了提高沟通成功率的效果。
在本实施例的可选实施方式中，对于本实施例步骤S204中涉及到的对对话内容中的 每条消息进行单独情感分析以及对对话内容进行整体情感分析，得到分析结果的方式，可以通过如下方式来实现：
步骤S204-11，基于文本分析模型识别对话内容中的每条消息对应的意图；
其中，该文本分析模型可以是通过关键字或者是整句话为模型来确定其意图，该意图对于服务对象的第二用户可以包括：打招呼、常规问题咨询、问题描述、常规回答客服问题、请求客服操作、礼貌语、不耐烦问题咨询、态度不端正回应等等；而该意图对应的对于提供服务的第一用户可以包括：欢迎语、解答客服问题、引导客服操作、问客户问题、礼貌语、结束语、态度不端正回应等等。需要说明的是，上述意图仅仅是举例说明，还可以包括其他意图，上述举例说明并不构成对本申请的限定。
步骤S204-12，根据识别出的每条消息对应的意图进行情感分析，得到第一分析结果，以及基于该第一分析结果对所述对话内容进行整体情感分析，得到所述第二分析结果。
其中，以网购平台中消费者作为第二用户，商家客服作为第一用户为例，其对话内容为意图为问题咨询与解答客服问题，如消费者问：“平时穿34的裤子，选哪个码子？”；客服回：“请问身高是多少厘米？”可见，这段对话内容中消费者问了问题，客服针对问题进行了回应，没有其他客套或增加好感的语气词。因此，对这段对话的每条消息进行情感分析，得到的每条消息的第一分析结果均为中性，即每条消息的情感程度均为中性，从而得到对话内容的整体情感分析的第二分析结果也为中性。当然，也有情感为正面的或者是负面的，例如正面的是，对话内容为意图为问题咨询与解答客服问题，消费者问：“麻烦明早帮我发快递”，客服回“可以的，亲”。在该问题咨询与解答客服问题中，客服对于消费者的问题进行积极回复，并附带上了带有尊称的“亲”，从而使得消费者对客服的回答比较满意。负面的是，对话内容为请求客服操作与回应操作结论，消费者在连发了几条信息都没有收到回应后：“是不是连信息都懒得看”，客服回应：“是的”。在该消息对应的意图中分析出消费者的不满和客服的服务态度差的结论。因此，基于每条消息的情感分析得到第一分析结果，进而可以基于该第一分析结果得到对话内容的整体的情感分析结果，例如，根据顾客前面每条消息进行情感分析的第一分析结果为情感倾向均是负面，经过客服处理，后续顾客的情感偏正面，那么整段对话的客户的情感也是偏正面的，即第二分析结果为正面的。但是如果客户的情绪始终是负面的，没有发生转变，那么整段对话客户的情感也还是负面的，即第二分析结果为负面的。
需要说明的是，对于本实施例中涉及到的文本分析模型可以通过训练得到，也就是说，在本实施例的步骤S204-11的在基于文本分析模型识别对话内容中的每条消息对应 的意图之前，本实施例的方法步骤还可以包括：
步骤S208，采用样本数据对文本分析模型进行训练，输出预判意图标签，其中，样本数据包括以下之一：与第一用户关联的消息样本、与第二用户关联的消息样本；
其中，对于上述样本数据中的与第一用户关联的消息样本和与第二用户关联的消息样本，在具体实施方式中可以是，例如，与第一用户关联的消息样本“亲，您报一下您的身高、体重”、“亲，我们的产品是百分百纯棉的”、“本店支持七天无理由退货的”、“当天下午三点前付款的当天发货”；与第二用户关联的消息样本“付款后什么时候能够发货？”、“你们的尺码标准码？”、“支持退货的吧？”。当然上述消息样本仅仅是举例说明，还可以是其他消息样本，样本越多越好，这其中最好是不同的情感程度都具有比较均衡的比例，例如将情感程度划分为：负面、中性、正面；对于这三种情感程度的样本尽量都是均很覆盖，这样才能保证输出的预判意图标签比较合理，进而以便后续情感分析。需要说明的是，该情感程度的划分也仅仅是举例说明，也可以根据实际请将情感程度更加细分的划分，例如负面可以划分为：非常负面和一般负面；正面可以划分为一般正面和非常正面。也就是说，情感程度的划分可以根据实际情感进行相应的设置和调整，上述仅仅是举例说明。
步骤S210，将预判意图标签与目标意图标签进行比对，得到比对结果；
步骤S212，根据比对结果调整文本分析模型所使用的模型参数，并重新采用样本数据对调整后的文本分析模型进行重复训练，直至预判意图标签与目标意图标签相匹配。
其中，是由于预判意图标签与目标意图标签不相匹配或者是不完全相匹配，则需要调整文本分析模型所使用的模型参数，即该模型参数能够对样本数据进行调整，进而对调整后的样本数据进行重复训练，直至预判意图标签与目标意图标签相匹配。也就是说，通过步骤S210和步骤S212可以将文本分析模型中的意图标签与目标意图标签相匹配，得到最终想要的文本分析模型。
在本实施例的另一个可选实施方式中，对于本实施例中上述步骤S204-11涉及到的基于文本分析模型识别对话内容中的每条消息对应的意图的方式可以是包括：
步骤S204-111，确定对话内容中的每条消息的消息来源；
其中，消息的来源就是指在对话内容该消息是由谁发出的，即是由第一用户(客服)发出的还是由第二用户(消费者/客户)发出的，例如在一段对话中：
客户：什么时候发货？
客户：发什么快递？
客服：在的哦，亲。
客服：亲，快递默认A快递、B快递、以及C快递；暂时没有其他快递哦，也不接受指定快递哦，麻烦您谅解哈。
客户：今天能不能发货？
客服：亲，这边会尽快给您安排发出的哦；
客户：大概什么时候发？
客户：不要尽快！！
客服：亲，具体时间需要看仓库那边的安排哦。
客服：仓库那边会加紧安排处理的哦。
客户：嗯嗯，麻烦尽快吧。
客服：好的哦。
可见，在上述对话中“什么时候发货”、“发什么快递？”、“大概什么时候发？”、“不要尽快！！”、“嗯嗯，麻烦尽快吧”，这些都是客户发的。而“在的哦，亲”、“亲，快递默认申通、圆通、邮政暂时没有其他快递哦，也不接受指定快递哦，麻烦您谅解哈”、“亲，这边会尽快给您安排发出的哦”、“亲，具体时间需要看仓库那边的安排哦”“仓库那边会加紧安排处理的哦”、“好的哦”，这些都是由客服回应或发出的。
步骤S204-112，将对话内容中的每条消息输入至与消息来源对应的文本分析模型，输出预判意图标签；
步骤S204-113，根据预判意图标签识别对话内容中的每条消息对应的意图。
其中，对于步骤S204-112和步骤S204-113，在确定了每条消息的来源之后，才能进一步输出该消息的预判意图标签，进而确定每条消息对应的意图。还是以上述步骤S204-111中对话为例，对于客户发出的“什么时候发货”、“发什么快递？”输入至与消息来源对应的文本分析模型，得到的意图是问题咨询；而将顾客发出的是“大概什么时候发？”、“不要尽快！！”输入对应的文本分析模型，则得到的意图是不耐烦。而将客服回复的“亲，快递默认A快递、B快递、以及C快递，暂时没有其他快递哦，也不接受指定快递哦，麻烦您谅解哈”、“亲，这边会尽快给您安排发出的哦”输入至对应的文本分析模型，则得到的意图是解答客户问题；而将客服回复的“亲，这边会尽快给您安排发出的哦”、“亲，具体时间需要看仓库那边的安排哦”“仓库那边会加紧安排处理的哦”输入至对应的文本分析模型，则得到的意图为礼貌回复客户。也就是说， 通过上述步骤可以确定对话内容中的每一条消息的意图。
在本实施例的另一个可选实施方式中，对于本实施例步骤S204-12中涉及到的根据识别出的每条消息对应的意图进行单独情感分析以及对对话内容进行整体情感分析，得到分析结果的方式，可以通过如下方式来实现：
步骤S204-121，根据识别出的每条消息对应的意图从情感多标签分类关系中确定关联的情感标签，其中，情感多标签分类关系用于维护每条消息对应的意图标签与情感标签之间的映射关系，情感多标签分类包括：至少一个层级的负面情感标签、中性情感标签、至少一个层级的正面情感标签；
其中，基于上述本实施例中的分析可知，在本实施例中意图对于服务对象的第二用户可以包括：打招呼、常规问题咨询、问题描述、常规回答客服问题、请求客服操作、礼貌语、不耐烦咨询、态度不端正回应等等；而该意图对应的对于提供服务的第一用户可以包括：欢迎语、解答客服问题、引导客服操作、问客户问题、礼貌语、结束语、态度不端正回应等等。
以上述S204-111中对话为例，根据该情感多标签分类关系来确定关联的情感标签；
客户：什么时候发货？(常规问题咨询，因此对应于中性情感标签)
客户：发什么快递？(常规问题咨询，因此对应于中性情感标签)
客服：在的哦，亲。(常规欢迎语，因此对应于中性情感标签)
客服：亲，快递默认申通、圆通、邮政暂时没有其他快递哦，也不接受指定快递哦，麻烦您谅解哈。(常规解答客户问题，因此对应于中性情感标签)
客户：今天能不能发货？(常规问题咨询，因此对应于中性情感标签)
客服：亲，这边会尽快给您安排发出的哦；(常规解答客服问题，因此对应于中性情感标签)
客户：大概什么时候发？，不要尽快！！(不耐烦咨询、带有一定的负面情绪，因此是负面情感标签)
客服：亲，具体时间需要看仓库那边的安排哦。(积极回应客户，正面情感标签)
客服：仓库那边会加紧安排处理的哦。(积极回应客户，正面情感标签)
客户：嗯嗯，麻烦尽快吧。(礼貌回应，正面情感标签)
客服：好的哦。(结束用语，正面情感标签)
由上述对话内容可知，可以根据每条消息对应的意图确定关联的情感标签。在本实施例中负面情感标签可以包括一般负面情感标签和非常负面情感标签；正面情感标签可 以包括：一般正面情感标签和非常正面情感标签。当然还可以是将情感标签更加的细化的划分，上述仅仅是举例说明。
步骤S204-122，根据情感标签对应的情感程度确定第一分析结果和第二分析结果。
以上述步骤S204-122中的对话内容为例，根据对话内容中每条消息的情感标标签能够确定每一条消息的情感程度，而基于每一条消息的情感标签从整体上可以确定客户(第二用户)和客服(第一用户)的整体情感，在上述对话内容中，客服由于体现了不耐烦、着急的负面情绪，因此整体上该客户的整体情绪为负面的，而客服一直是积极回应，因此该客服的整体情感是正面的。
下述对话内容则是客户和客服整体情感均为负面的对话。
客户：这个用电量大吗？(常规问题咨询，中性情感标签)
客户：跟风扇比起来哪个用电量大？(常规问题咨询，中性情感标签)
客服：一小时0.075度电。(常规解答客户问题，中性情感标签)
客户：风扇呢？(常规问题咨询，中性情感标签)
客服：我怎么知道你说的风扇是什么风扇，也不知道多少功率的。(不耐烦解答客户问题，负面情感标签)
客户：渍渍！！你这态度绝了！！(不耐烦回应，负面情感标签)
客服：你不觉得你的问题很奇怪呢？问个“风扇呢”，风扇多了去了。(不耐烦解答客户问题，负面情感标签)
客户：你不知道说个大约的？(不耐烦回应，负面情感标签)
客服：？？？？(消极回应，负面情感标签)
可见，上述双方在常规咨询与常规回答后，均进入到态度不端正的对话中，因此，两者整体情感均是负面的。
在本实施例的再一个可选实施方式中，本实施例步骤S206中涉及到的根据分析结果分别确定第一用户与第二用户在对话内容中的情感变化趋势的方式，可以通过如下方式来实现：
步骤S206-11，通过第一分析结果对对话内容中的每条消息对应的情感程度进行统计，得到第一统计结果，通过第二分析结果对所述对话内容按照时间进行的情感变化进行统计，得到第二统计结果；
其中，对于上述统计的方式首先是要将整个对话内容每条消息的情感程度进行统计，即每一条消息都有对应的情感标签得到第一统计结果。然后，就是基于整个对话内容， 对每个情感标签产生的时间节点进行标记，也就是说，需要每个情感标签是在对话内容哪个时间段产生的，进而可以根据该第一分析结果以及对话内容的时间进度情感变化进行统计，得到第二统计结果。
步骤S206-12，至少根据统第一统计结果和第二统计结果分别确定第一用户与第二用户在对话内容中的情感变化趋势。
其中，根据第一统计结果和第二统计结果确定第一用户与第二用户的情感变化趋势，例如，根据第一统计结果和第二统计结果，当前第二用户的整体情绪是负面的，接下来第二用户的整体情绪依然可能会是负面的，因此，第一用户可以根据该第二用户的整体情绪变化趋势来改变回应内容。在具体应用场景中，客户一开始的倾向是负面，经过客服处理，偏正面，那么整段对话的客户的情感也是偏正面的。
需要说明的是，对于前述的各方法实施例，为了简单描述，故将其都表述为一系列的动作组合，但是本领域技术人员应该知悉，本发明并不受所描述的动作顺序的限制，因为依据本发明，某些步骤可以采用其他顺序或者同时进行。其次，本领域技术人员也应该知悉，说明书中所描述的实施例均属于优选实施例，所涉及的动作和模块并不一定是本发明所必须的。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到根据上述实施例的方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本发明的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，或者网络设备等)执行本发明各个实施例所述的方法。
实施例2
根据本发明实施例，还提供了一种用于实施上述对话情感分析方法的对话情感分析装置，图3是根据本发明实施例的对话情感分析装置的结构示意图，如图3所示，该装置包括：
获取模块32，用于获取第一用户与第二用户之间的对话内容，其中，第一用户为服务提供者，第二用户为服务对象；
需要说明的是，由于本实施例中的第一用户为服务提供者，第二用户为服务对象；因此，在具体应用场景中，该第一用户可以是网购平台某一商家的客服，而第二用户则为咨询该商家商品的网购用户。如果该商家的商品是衣服的话，则该第一用户与第二用 户之间的对话内容可以是：第二用户咨询的某一件衣服的材质、大小、折扣等等，以及第一用户根据第二用户咨询的问题所回复的相关内容。当然，上述第一用户和第二用户并不局限于是网购平台的用户，也可以是其他咨询服务类平台的用户。当然，上述网购的具体应用场景仅仅是举例说明，也可以是其他能够场景，例如，电信服务业中客服与机主之间的对话内容，也可以是咨询服务业中的客服与顾客之间的对话内容，或者是其他场景中的对话内容等等。只要是存在两者之间的对话内容，则可以应用本申请进行情感倾向的分析。
分析模块34，与获取模块32耦合连接，用于对对话内容中的每条消息进行单独情感分析得到第一分析结果，以及基于每条消息情感分析的第一分析结果对对话内容进行整体情感分析，得到第二分析分析结果；
其中，以网购平台中消费者作为第二用户，商家客服作为第一用户为例，其对话内容为，消费者问：“平时穿34的裤子，选哪个码子？”；客服回：“请问身高是多少厘米？”可见，这段对话内容中消费者问了问题，客服针对问题进行了回应，没有其他客套或增加好感的语气词。因此，对这段对话的每条消息进行单独情感分析，得到的第一分析结果为中性，即每条消息的情感程度均为中性。当然，也有情感为正面的或者是负面的，例如正面的是，消费者问：“麻烦明早帮我发快递”。客服回“可以的，亲”。负面的是，消费者在连发了几条信息都没有收到回应后：“是不是连信息都懒得看”，客服回应：“是的”。而对于一段对话内容有正面，负面、中性的情感程度的情况下，可以根据具体每条消息的情感分析得到的第一分析结果得到用于表征整体的情感倾向的第二分析结果，例如，在网购的退货对话内容中，顾客一开始的语气与谈话内容都是不友善的，用词也比较尖锐，但是客服的会话一致是非常客气的，在客服的友好的态度下，顾客的态度也慢慢变好了，最终和平的退了货，并选择了其他商品来代替。可见，对该顾客第一分析结果为情感由负面到中性直到最后的正面，这样就得到顾客的整体情感分析的第二分析结果为正面，进而得到整个对话过程中的情感变化趋势。
确定模块36，与分析模块34耦合连接，用于至少根据第一分析结果和第二分析结果分别确定第一用户与第二用户在对话内容中的情感变化趋势。
其中，该情感变化趋势可以是一开始是中性的，后续沟通过程中可能由于客服回复不及时或者客服态度不好，则对对话内容的中的消息分析得到此时的对话内容为负面。也就是说，根据前面对话内容，第二用户的情感由中性变为负面。客服看到这个情感变化之后，得到的结论是如果继续不理会或态度不好则第二用户的情感将会以更加负面的 趋势演化。因此，根据该情感变化趋势，客服可以以端正态度或积极回复消息，并对问题进行耐心解释的方式进行回复，从而使得双方的对话内容回到了正面结果。
此处需要说明的是，上述获取模块32、分析模块34和确定模块36对应于实施例1中的步骤S202至步骤S206，两个模块与对应的步骤所实现的实例和应用场景相同，但不限于上述实施例一所公开的内容。需要说明的是，上述模块作为装置的一部分可以运行在实施例一提供的计算机终端10中。
可选地，本实施例中的分析模块34可以包括：识别单元，与分析单元耦合连接，用于基于文本分析模型识别对话内容中的每条消息对应的意图；
其中，该识别单元可以基于文本分析模型的关键字或者是整句话为模型来确定其意图，该意图对于服务对象的第二用户可以包括：打招呼、常规问题咨询、问题描述、常规回答客服问题、请求客服操作、礼貌语、不耐烦问题咨询、态度不端正回应等等；而该意图对应的对于提供服务的第一用户可以包括：欢迎语、解答客服问题、引导客服操作、问客户问题、礼貌语、结束语、态度不端正回应等等。需要说明的是，上述意图仅仅是举例说明，还可以包括其他意图，上述举例说明并不构成对本申请的限定。
分析单元，用于根据识别出的每条消息对应的意图进行情感分析，得到第一分析结果，以及基于第一分析结果对对话内容进行整体情感分析，得到所述第二分析结果。
其中，以网购平台中消费者作为第二用户，商家客服作为第一用户为例，其对话内容为意图为问题咨询与解答客服问题，如消费者问：“平时穿34的裤子，选哪个码子？”；客服回：“请问身高是多少厘米？”可见，这段对话内容中消费者问了问题，客服针对问题进行了回应，没有其他客套或增加好感的语气词。因此，该分析单元对这段对话的每条消息进行情感分析，得到的每条消息的第一分析结果均为中性，即每条消息的情感程度均为中性，从而得到对话内容的整体情感分析的第二分析结果也为中性。当然，也有情感为正面的或者是负面的，例如正面的是，对话内容为意图为问题咨询与解答客服问题，消费者问：“麻烦明早帮我发快递”，客服回“可以的，亲”。在该问题咨询与解答客服问题中，客服对于消费者的问题进行积极回复，并附带上了带有尊称的“亲”，从而使得消费者对客服的回答比较满意。负面的是，对话内容为请求客服操作与回应操作结论，消费者在连发了几条信息都没有收到回应后：“是不是连信息都懒得看”，客服回应：“是的”。在该消息对应的意图中分析出消费者的不满和客服的服务态度差的结论。因此，基于每条消息的情感分析得到第一分析结果，进而可以基于该第一分析结果得到对话内容的整体的情感分析结果，例如，根据顾客前面每条消息进行情感分析的 第一分析结果为情感倾向均是负面，经过客服处理，后续顾客的情感偏正面，那么整段对话的客户的情感也是偏正面的，即第二分析结果为正面的。但是如果客户的情绪始终是负面的，没有发生转变，那么整段对话客户的情感也还是负面的，即第二分析结果为负面的。
图4是根据本发明实施例的对话情感分析装置的可选结构示意图，如图4所示，装置还包括：
第一训练模块42，与比对模块44耦合连接，用于在基于文本分析模型识别对话内容中的每条消息对应的意图之前，采用样本数据对文本分析模型进行训练，输出预判意图标签，其中，样本数据包括以下之一：与第一用户关联的消息样本、与第二用户关联的消息样本；
其中，对于上述样本数据中的与第一用户关联的消息样本和与第二用户关联的消息样本，在具体实施方式中可以是，例如，与第一用户关联的消息样本“亲，您报一下您的身高、体重”、“亲，我们的产品是百分百纯棉的”、“本店支持七天无理由退货的”、“当天下午三点前付款的当天发货”；与第二用户关联的消息样本“付款后什么时候能够发货？”、“你们的尺码标准码？”、“支持退货的吧？”。当然上述消息样本仅仅是举例说明，还可以是其他消息样本，样本越多越好，这其中最好是不同的情感程度都具有比较均衡的比例，例如将情感程度划分为：负面、中性、正面；对于这三种情感程度的样本尽量都是均很覆盖，这样才能保证输出的预判意图标签比较合理，进而以便后续情感分析。需要说明的是，该情感程度的划分也仅仅是举例说明，也可以根据实际请将情感程度更加细分的划分，例如负面可以划分为：非常负面和一般负面；正面可以划分为一般正面和非常正面。也就是说，情感程度的划分可以根据实际情感进行相应的设置和调整，上述仅仅是举例说明。
比对模块44，与第二训练模块46耦合连接，用于将预判意图标签与目标意图标签进行比对，得到比对结果；
第二训练模块46，与分析模块34耦合连接，用于根据比对结果调整文本分析模型所使用的模型参数，并重新采用样本数据对调整后的文本分析模型进行重复训练，直至预判意图标签与目标意图标签相匹配。
可选地，本实施例中涉及到的识别单元包括：
第一确定子单元，用于确定对话内容中的每条消息的消息来源；
其中，消息的来源就是指在对话内容该消息是由谁发出的，即是由第一用户(客服) 发出的还是由第二用户(消费者/客户)发出的，例如在一段对话中：
客户：什么时候发货？
客户：发什么快递？
客服：在的哦，亲。
客服：亲，快递默认A快递、B快递、以及C快递，暂时没有其他快递哦，也不接受指定快递哦，麻烦您谅解哈。
客户：今天能不能发货？
客服：亲，这边会尽快给您安排发出的哦；
客户：大概什么时候发？
客户：不要尽快！！
客服：亲，具体时间需要看仓库那边的安排哦。
客服：仓库那边会加紧安排处理的哦。
客户：嗯嗯，麻烦尽快吧。
客服：好的哦。
可见，第一确定子单元可以确定在上述对话中“什么时候发货”、“发什么快递？”、“大概什么时候发？”、“不要尽快！！”、“嗯嗯，麻烦尽快吧”是客户发的。而第一确定子单元确定“在的哦，亲”、“亲，快递默认申通、圆通、邮政暂时没有其他快递哦，也不接受指定快递哦，麻烦您谅解哈”、“亲，这边会尽快给您安排发出的哦”、“亲，具体时间需要看仓库那边的安排哦”“仓库那边会加紧安排处理的哦”、“好的哦”，这些都是由客服回应或发出的。
输入子单元，与第一确定子单元耦合连接，用于将对话内容中的每条消息输入至与消息来源对应的文本分析模型，输出预判意图标签；
识别子单元，与输入子单元耦合连接，用于根据预判意图标签识别对话内容中的每条消息对应的意图。
其中，在第一确定子单元确定了每条消息的来源之后，才能进一步由输入子单元输出该消息的预判意图标签，进而识别子单元确定每条消息对应的意图。还是以上述对话为例，对于客户发出的“什么时候发货”、“发什么快递？”输入至与消息来源对应的文本分析模型，得到的意图是问题咨询；而将顾客发出的“大概什么时候发？”、“不要尽快！！”输入对应的文本分析模型，则得到的意图是不耐烦。而将客服汇入的“亲，快递默认A快递、B快递、以及C快递，暂时没有其他快递哦，也不接受指定快递哦， 麻烦您谅解哈”、“亲，这边会尽快给您安排发出的哦”输入至对应的文本分析模型，则得到的意图是解答客户问题；而将客服回复的“亲，这边会尽快给您安排发出的哦”、“亲，具体时间需要看仓库那边的安排哦”“仓库那边会加紧安排处理的哦”输入至对应的文本分析模型，则得到的意图为礼貌回复客户。也就是说，通过上述步骤可以确定对话内容中的每一条消息的意图。
可选地，本实施例中的分析单元包括：
第一确定子单元，用于根据识别出的每条消息对应的意图从情感多标签分类关系中确定关联的情感标签，其中，情感多标签分类关系用于维护每条消息对应的意图标签与情感标签之间的映射关系，情感多标签分类包括：至少一个层级的负面情感标签、中性情感标签、至少一个层级的正面情感标签；
其中，基于上述本实施例中的分析可知，在本实施例中意图对于服务对象的第二用户可以包括：打招呼、常规问题咨询、问题描述、常规回答客服问题、请求客服操作、礼貌语、不耐烦咨询、态度不端正回应等等；而该意图对应的对于提供服务的第一用户可以包括：欢迎语、解答客服问题、引导客服操作、问客户问题、礼貌语、结束语、态度不端正回应等等。
以上述对话为例，该第一确定子单元根据该情感多标签分类关系来确定关联的情感标签；
客户：什么时候发货？(常规问题咨询，因此对应于中性情感标签)
客户：发什么快递？(常规问题咨询，因此对应于中性情感标签)
客服：在的哦，亲。(常规欢迎语，因此对应于中性情感标签)
客服：亲，快递默认申通、圆通、邮政暂时没有其他快递哦，也不接受指定快递哦，麻烦您谅解哈。(常规解答客户问题，因此对应于中性情感标签)
客户：今天能不能发货？(常规问题咨询，因此对应于中性情感标签)
客服：亲，这边会尽快给您安排发出的哦；(常规解答客服问题，因此对应于中性情感标签)
客户：大概什么时候发？，不要尽快！！(不耐烦咨询、带有一定的负面情绪，因此是负面情感标签)
客服：亲，具体时间需要看仓库那边的安排哦。(积极回应客户，正面情感标签)
客服：仓库那边会加紧安排处理的哦。(积极回应客户，正面情感标签)
客户：嗯嗯，麻烦尽快吧。(礼貌回应，正面情感标签)
客服：好的哦。(结束用语，正面情感标签)
由上述对话内容可知，可以根据每条消息对应的意图确定关联的情感标签。在本实施例中负面情感标签可以包括一般负面情感标签和非常负面情感标签；正面情感标签可以包括：一般正面情感标签和非常正面情感标签。当然还可以是将情感标签更加的细化的划分，上述仅仅是举例说明。
第三确定子单元，用于根据情感标签对应的情感程度确定第一分析结果和第二分析结果。
其中，以上述对话内容为例，该第三确定子单元根据对话内容中每条消息的情感标标签能够确定每一条消息的情感程度，而基于每一条消息的情感标签从整体上可以确定客户(第二用户)和客服(第一用户)的整体情感，在上述对话内容中，客服由于体现了不耐烦、着急的负面情绪，因此整体上该客户的情绪为负面的，而客服一直是积极回应，因此该客服的情绪是正面的。
下述对话内容则是客户和客服整体情感均为负面的对话。
客户：这个用电量大吗？(常规问题咨询，中性情感标签)
客户：跟风扇比起来哪个用电量大？(常规问题咨询，中性情感标签)
客服：一小时0.075度电。(常规解答客户问题，中性情感标签)
客户：风扇呢？(常规问题咨询，中性情感标签)
客服：我怎么知道你说的风扇是什么风扇，也不知道多少功率的。(不耐烦解答客户问题，负面情感标签)
客户：渍渍！！你这态度绝了！！(不耐烦回应，负面情感标签)
客服：你不觉得你的问题很奇怪呢？问个“风扇呢”，风扇多了去了。(不耐烦解答客户问题，负面情感标签)
客户：你不知道说个大约的？(不耐烦回应，负面情感标签)
客服：？？？？(消极回应，负面情感标签)
可见，上述双方在常规咨询与常规回答后，均进入到态度不端正的对话中，因此，两者整体情感均是负面的。
可选地，本实施例中涉及到的确定模块36包括：
统计单元，用于通过第一分析结果对对话内容中的每条消息对应的情感程度进行统计，得到第一统计结果，通过第二分析结果对所述对话内容按照时间进行的情感变化进行统计，得到第二统计结果；
其中，该统计单元可以是要将整个对话内容每条消息的情感程度进行统计，即每一条消息都有对应的情感标签得到第一统计结果。然后，就是基于整个对话内容，对每个情感标签产生的时间节点进行标记，也就是说，需要每个情感标签是在对话内容哪个时间段产生的，进而可以根据该第一分析结果以及对话内容的时间进度情感变化进行统计，得到第二统计结果。
确定单元，用于至少根据第一统计结果和第二统计结果分别确定第一用户与第二用户在对话内容中的情感变化趋势。
其中，根据第一统计结果和第二统计结果确定第一用户与第二用户的情感变化趋势，例如，根据第一统计结果和第二统计结果，当前第二用户的整体情绪是负面的，接下来第二用户的整体情绪依然可能会使负面的，因此，第一用户可以根据该第二用户的整体情绪变化趋势来改变回应内容。在具体应用场景中，客户一开始的倾向是负面，经过客服处理，偏正面，那么整段对话的客户的情感也是偏正面的。
实施例3
本发明的实施例可以提供一种处理器，该处理器可以用于计算机终端，该计算机终端可以是计算机终端群中的任意一个计算机终端设备。可选地，在本实施例中，上述计算机终端也可以替换为移动终端等终端设备。
可选地，在本实施例中，上述计算机终端可以位于计算机网络的多个网络设备中的至少一个网络设备。
可选地，图1是根据本发明实施例的一种计算机终端的结构框图。如图1所示，该计算机终端A可以包括：一个或多个处理器、存储器、以及接口。
其中，存储器可用于存储软件程序以及模块，如本发明实施例中的安全漏洞检测方法和装置对应的程序指令/模块，处理器通过运行存储在存储器内的软件程序以及模块，从而执行各种功能应用以及数据处理，即实现上述的系统漏洞攻击的检测方法。存储器可包括高速随机存储器，还可以包括非易失性存储器，如一个或者多个磁性存储装置、闪存、或者其他非易失性固态存储器。在一些实例中，存储器可进一步包括相对于处理器远程设置的存储器，这些远程存储器可以通过网络连接至终端A。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
处理器可以通过传输装置调用存储器存储的信息及应用程序，以执行下述步骤：
S1，获取第一用户与第二用户之间的对话内容，其中，所述第一用户为服务提供者，所述第二用户为服务对象；
S2，对所述对话内容中的每条消息进行单独情感分析，得到第一分析结果，以及基于每条消息情感分析的第一结果对对话内容进行整体情感分析，得到第二分析结果；
S3，至少根据第一分析结果和第二分析结果分别确定所述第一用户与第二用户在所述对话内容中的情感变化趋势。
可选的，上述处理器还可以执行如下步骤的程序代码：基于文本分析模型识别所述对话内容中的每条消息对应的意图；根据识别出的每条消息对应的意图进行情感分析，得到第一分析结果，以及基于第一分析结果对对话内容进行整体情感分析，得到第二分析结果。
可选的，上述处理器还可以执行如下步骤的程序代码：在基于文本分析模型识别对话内容中的每条消息对应的意图之前，采用样本数据对文本分析模型进行训练，输出预判意图标签，其中，样本数据包括以下之一：与第一用户关联的消息样本、与第二用户关联的消息样本；将预判意图标签与目标意图标签进行比对，得到比对结果；根据比对结果调整文本分析模型所使用的模型参数，并重新采用样本数据对调整后的文本分析模型进行重复训练，直至预判意图标签与目标意图标签相匹配。
可选的，上述处理器还可以执行如下步骤的程序代码：确定对话内容中的每条消息的消息来源；将对话内容中的每条消息输入至与消息来源对应的文本分析模型，输出预判意图标签；根据预判意图标签识别对话内容中的每条消息对应的意图。
可选的，上述处理器还可以执行如下步骤的程序代码：根据识别出的每条消息对应的意图从情感多标签分类关系中确定关联的情感标签，其中，情感多标签分类关系用于维护每条消息对应的意图标签与情感标签之间的映射关系，情感多标签分类包括：至少一个层级的负面情感标签、中性情感标签、至少一个层级的正面情感标签；根据情感标签对应的情感程度确定第一分析结果和第二分析结果。
可选的，上述处理器还可以执行如下步骤的程序代码：通过第一分析结果对对话内容中的每条消息对应的情感程度进行统计，得到第一统计结果，通过第二分析结果对对话内容按照时间进度的情感变化进行统计，得到第二统计结果；至少根据第一统计结果和第二统计结果分别确定第一用户与第二用户在对话内容中的情感变化趋势。
采用本发明实施例，提供了一种对话情感分析的方案。能够对服务提供者的第一用户与服务对象的第二用户的对话内容中的每一条消息进行情感分析得到第一分析结果以及基于该第一分析结果对对话内容进行整体情感分析得到第二分析结果，根据该第一分析结果和第二分析结果来判断第一用户和第二用户在对话内容中的情感变化趋势，能够 快速确定双方的意向、态度、满意度等等，以使得双方根据该情感变化趋势调整后续对话内容，进而解决了相关技术中对于消费者与客服的对话内容的情感倾向，客服自己的理解和判断存在主观臆断的技术问题，达到了提高沟通成功率的效果。
本领域普通技术人员可以理解，图10所示的结构仅为示意，计算机终端也可以是智能手机(如Android手机、iOS手机等)、平板电脑、掌声电脑以及移动互联网设备(Mobile Internet Devices，MID)、PAD等终端设备。图10其并不对上述电子装置的结构造成限定。例如，计算机终端10还可包括比图10中所示更多或者更少的组件(如网络接口、显示装置等)，或者具有与图10所示不同的配置。
本领域普通技术人员可以理解上述实施例的各种方法中的全部或部分步骤是可以通过程序来指令终端设备相关的硬件来完成，该程序可以存储于一计算机可读存储介质中，存储介质可以包括：闪存盘、只读存储器(Read-Only Memory，ROM)、随机存取器(Random Access Memory，RAM)、磁盘或光盘等。
实施例4
本发明的实施例还提供了一种存储介质。可选地，在本实施例中，上述存储介质可以用于保存上述实施例一所提供的对话情感分析的方法所执行的程序代码。
可选地，在本实施例中，上述存储介质可以位于计算机网络中计算机终端群中的任意一个计算机终端中，或者位于移动终端群中的任意一个移动终端中。
可选地，在本实施例中，存储介质被设置为存储用于执行以下步骤的程序代码：
S1，获取第一用户与第二用户之间的对话内容，其中，第一用户为服务提供者，第二用户为服务对象；
S2，对对话内容中的每条消息进行单独情感分析，得到第一分析结果，以及基于每条消息情感分析的第一分析结果对对话内容进行整体情感分析，得到第二分析结果；
S3，至少根据第一分析结果和第二分析结果分别确定第一用户与第二用户在对话内容中的情感变化趋势。
可选的，上述存储介质还可以执行如下步骤的程序代码：基于文本分析模型识别对话内容中的每条消息对应的意图；根据识别出的每条消息对应的意图进行情感分析，得到第一分析结果，以及基于所述第一分析结果对所述对话内容进行整体情感分析，得到所述第二分析结果。
可选的，上述存储介质还可以执行如下步骤的程序代码：在基于文本分析模型识别对话内容中的每条消息对应的意图之前，采用样本数据对文本分析模型进行训练，输出 预判意图标签，其中，样本数据包括以下之一：与第一用户关联的消息样本、与第二用户关联的消息样本；将预判意图标签与目标意图标签进行比对，得到比对结果；根据比对结果调整文本分析模型所使用的模型参数，并重新采用样本数据对调整后的文本分析模型进行重复训练，直至预判意图标签与目标意图标签相匹配。
可选的，上述存储介质还可以执行如下步骤的程序代码：确定对话内容中的每条消息的消息来源；将对话内容中的每条消息输入至与消息来源对应的文本分析模型，输出预判意图标签；根据预判意图标签识别对话内容中的每条消息对应的意图。
可选的，上述存储介质还可以执行如下步骤的程序代码：根据识别出的每条消息对应的意图从情感多标签分类关系中确定关联的情感标签，其中，情感多标签分类关系用于维护每条消息对应的意图标签与情感标签之间的映射关系，情感多标签分类包括：至少一个层级的负面情感标签、中性情感标签、至少一个层级的正面情感标签；根据情感标签对应的情感程度确定第一分析结果和第二分析结果。
可选的，上述存储介质还可以执行如下步骤的程序代码：通过第一分析结果对对话内容中的每条消息对应的情感程度进行统计，得到第一统计结果，通过第二分析结果对所述对话内容按照时间进度的情感变化进行统计，得到第二统计结果；至少根据第一统计结果和第二统计结果分别确定第一用户与第二用户在对话内容中的情感变化趋势。
上述本发明实施例序号仅仅为了描述，不代表实施例的优劣。
在本发明的上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述的部分，可以参见其他实施例的相关描述。
在本申请所提供的几个实施例中，应该理解到，所揭露的技术内容，可通过其它的方式实现。其中，以上所描述的装置实施例仅仅是示意性的，例如所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，单元或模块的间接耦合或通信连接，可以是电性或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本发明各个实施例中的各功能单元可以集成在一个处理单元中，也可以是 各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本发明的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可为个人计算机、服务器或者网络设备等)执行本发明各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、移动硬盘、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述仅是本发明的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本发明原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本发明的保护范围。

Claims (15)

  1. 一种对话情感分析方法，其特征在于，包括：
    获取第一用户与第二用户之间的对话内容，其中，所述第一用户为服务提供者，所述第二用户为服务对象；
    对所述对话内容中的每条消息进行单独情感分析，得到第一分析结果，以及基于每条消息情感分析的第一分析结果对所述对话内容进行整体情感分析，得到第二分析结果；
    至少根据所述第一分析结果和所述第二分析结果分别确定所述第一用户与所述第二用户在所述对话内容中的情感变化趋势。
  2. 根据权利要求1所述的方法，还包括：
    基于文本分析模型识别所述对话内容中的每条消息对应的意图；
    根据识别出的每条消息对应的意图进行情感分析，得到所述第一分析结果，以及基于所述第一分析结果对所述对话内容进行整体情感分析，得到所述第二分析结果。
  3. 根据权利要求2所述的方法，还包括：
    采用样本数据对所述文本分析模型进行训练，输出预判意图标签，其中，所述样本数据包括以下之一：与所述第一用户关联的消息样本、与所述第二用户关联的消息样本；
    将所述预判意图标签与目标意图标签进行比对，得到比对结果；
    根据所述比对结果调整所述文本分析模型所使用的模型参数，并重新采用所述样本数据对调整后的文本分析模型进行重复训练，直至所述预判意图标签与所述目标意图标签相匹配。
  4. 根据权利要求3所述的方法，还包括：
    确定所述对话内容中的每条消息的消息来源；
    将所述对话内容中的每条消息输入至与所述消息来源对应的文本分析模型，输出所述预判意图标签；
    根据所述预判意图标签识别所述对话内容中的每条消息对应的意图。
  5. 根据权利要求2所述的方法，还包括：
    根据识别出的每条消息对应的意图从情感多标签分类关系中确定关联的情感标签，其中，所述情感多标签分类关系用于维护每条消息对应的意图标签与情感标签之间的映射关系，情感多标签分类包括：至少一个层级的负面情感标签、中性情感标签、至少一个层级的正面情感标签；
    根据所述情感标签对应的情感程度确定所述第一分析结果和所述第二分析结果。
  6. 根据权利要求1所述的方法，其特征在于，根据所述分析结果分别确定所述第一用户与所述第二用户在所述对话内容中的情感变化趋势包括：
    通过所述第一分析结果对所述对话内容中的每条消息对应的情感程度进行统计，得到第一统计结果，通过所述第二分析结果对所述对话内容按照时间进度的情感变化进行统计，得到第二统计结果；
    至少根据所述第一统计结果和所述第二统计结果分别确定所述第一用户与所述第二用户在所述对话内容中的情感变化趋势。
  7. 根据权利要求1至6中任一项所述的方法，其特征在于，所述方法应用于电商领域。
  8. 一种对话情感分析装置，其特征在于，包括：
    获取模块，用于获取第一用户与第二用户之间的对话内容，其中，所述第一用户为服务提供者，所述第二用户为服务对象；
    分析模块，用于对所述对话内容中的每条消息进行单独情感分析，得到第一分析结果，以及基于每条消息情感分析的第一分析结果对所述对话内容进行整体情感分析，得到第二分析结果；
    确定模块，用于至少根据所述第一分析结果和所述第二分析结果分别确定所述第一用户与所述第二用户在所述对话内容中的情感变化趋势。
  9. 根据权利要求8所述的装置，所述分析模块包括：
    识别单元，用于基于文本分析模型识别所述对话内容中的每条消息对应的意图；
    分析单元，用于根据识别出的每条消息对应的意图进行情感分析，得到所述第一分析结果，以及基于所述第一分析结果对所述对话内容进行整体情感分析，得到所述第二分析结果。
  10. 根据权利要求9所述的装置，所述装置还包括：
    第一训练模块，用于采用样本数据对所述文本分析模型进行训练，输出预判意图标签，其中，所述样本数据包括以下之一：与所述第一用户关联的消息样本、与所述第二用户关联的消息样本；
    比对模块，用于将所述预判意图标签与目标意图标签进行比对，得到比对结果；
    第二训练模块，用于根据所述比对结果调整所述文本分析模型所使用的模型参数，并重新采用所述样本数据对调整后的文本分析模型进行重复训练，直至所述预判意图标 签与所述目标意图标签相匹配。
  11. 根据权利要求10所述的装置，所述识别单元包括：
    第一确定子单元，用于确定所述对话内容中的每条消息的消息来源；
    输入子单元，用于将所述对话内容中的每条消息输入至与所述消息来源对应的文本分析模型，输出所述预判意图标签；
    识别子单元，用于根据所述预判意图标签识别所述对话内容中的每条消息对应的意图。
  12. 根据权利要求9所述的装置，所述分析单元包括：
    第一确定子单元，用于根据识别出的每条消息对应的意图从情感多标签分类关系中确定关联的情感标签，其中，所述情感多标签分类关系用于维护每条消息对应的意图标签与情感标签之间的映射关系，情感多标签分类包括：至少一个层级的负面情感标签、中性情感标签、至少一个层级的正面情感标签；
    第三确定子单元，用于根据所述情感标签对应的情感程度确定所述第一分析结果和所述第二分析结果。
  13. 根据权利要求8所述的装置，所述确定模块包括：
    统计单元，用于通过所述第一分析结果对所述对话内容中的每条消息对应的情感程度进行统计，得到第一统计结果，通过所述第二分析结果对所述对话内容按照时间进行的情感变化进行统计，得到第二统计结果；
    确定单元，用于至少根据所述第一统计结果和所述第二统计结果分别确定所述第一用户与所述第二用户在所述对话内容中的情感变化趋势。
  14. 一种存储介质，其特征在于，所述存储介质包括存储的程序，其中，在所述程序运行时控制所述存储介质所在设备执行权利要求1至7中任意一项所述的方法。
  15. 一种处理器，其特征在于，所述处理器用于运行程序，其中，所述程序运行时执行权利要求1至7中任意一项所述的方法。
PCT/CN2019/123215 2018-12-11 2019-12-05 对话情感分析方法及装置、存储介质和处理器 WO2020119568A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811512090.7 2018-12-11
CN201811512090.7A CN111309899A (zh) 2018-12-11 2018-12-11 对话情感分析方法及装置、存储介质和处理器

Publications (1)

Publication Number Publication Date
WO2020119568A1 true WO2020119568A1 (zh) 2020-06-18

Family

ID=71077118

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/123215 WO2020119568A1 (zh) 2018-12-11 2019-12-05 对话情感分析方法及装置、存储介质和处理器

Country Status (2)

Country Link
CN (1) CN111309899A (zh)
WO (1) WO2020119568A1 (zh)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111859902A (zh) * 2020-07-16 2020-10-30 微医云(杭州)控股有限公司 一种文本处理方法、装置、设备及介质
CN111966805B (zh) * 2020-08-13 2021-11-09 贝壳找房(北京)科技有限公司 用于辅助实现会话的方法、装置、介质以及电子设备
CN112528017A (zh) * 2020-11-30 2021-03-19 北京搜狗科技发展有限公司 对话流自动分析方法及装置
CN115730070B (zh) * 2022-11-25 2023-08-08 重庆邮电大学 一种人机共情对话方法、系统、电子设备和介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106775665A (zh) * 2016-11-29 2017-05-31 竹间智能科技（上海）有限公司 基于情绪指标的情绪状态变化信息的获取方法及装置
CN107832294A (zh) * 2017-11-06 2018-03-23 广州杰赛科技股份有限公司 客服质量评价方法和装置
CN107895230A (zh) * 2017-11-06 2018-04-10 广州杰赛科技股份有限公司 客服质量评价方法和装置
CN108491519A (zh) * 2018-03-26 2018-09-04 上海智臻智能网络科技股份有限公司 人机交互方法及装置、存储介质、终端

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105830118A (zh) * 2013-08-10 2016-08-03 格林伊登美国控股有限责任公司 用于判定在线会话的结果和通过会话期间情绪表达的分析的类似谈话的方法和设备
CN107870896B (zh) * 2016-09-23 2021-06-25 深圳市云网万店电子商务有限公司 一种对话分析方法及装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106775665A (zh) * 2016-11-29 2017-05-31 竹间智能科技（上海）有限公司 基于情绪指标的情绪状态变化信息的获取方法及装置
CN107832294A (zh) * 2017-11-06 2018-03-23 广州杰赛科技股份有限公司 客服质量评价方法和装置
CN107895230A (zh) * 2017-11-06 2018-04-10 广州杰赛科技股份有限公司 客服质量评价方法和装置
CN108491519A (zh) * 2018-03-26 2018-09-04 上海智臻智能网络科技股份有限公司 人机交互方法及装置、存储介质、终端

Also Published As

Publication number Publication date
CN111309899A (zh) 2020-06-19

Similar Documents

Publication Publication Date Title
WO2020119568A1 (zh) 对话情感分析方法及装置、存储介质和处理器
US10063702B2 (en) Intelligent customer service systems, customer service robots, and methods for providing customer service
US20210224832A1 (en) Method and apparatus for predicting customer purchase intention, electronic device and medium
EP3246857A1 (en) Computer-implemented system and method for facilitating interactions via automatic agent responses
CN108681871B (zh) 一种提示信息的方法、终端设备及计算机可读存储介质
CN109816515A (zh) 基于互联网的信贷类产品选择系统及方法
US20230274280A1 (en) Dynamically populated user interface feature
US12047334B1 (en) Automated agent messaging system
US20160142344A1 (en) Automated response action in messaging application
CN111445263A (zh) 信息处理方法、装置和计算机系统及介质
US10891629B1 (en) Systems and methods for matching a customer with a service representative
CN112818689B (zh) 一种实体识别方法、模型训练方法及装置
US20210117972A1 (en) Rule-based messaging and electronic communication system
US11935111B2 (en) Voice-based transaction terminal ordering
CN115082134A (zh) 营销方法、装置、系统、设备及介质
JP6656216B2 (ja) コンテンツ配信方法、コンテンツ配信システム、コンテンツ配信サーバ及びコンテンツ配信プログラム
CN109002497B (zh) 一种基于群体评价关联用户的方法及装置
CN113643049A (zh) 一种网销客户管理方法及系统
US9954836B2 (en) Wireless network information flow conditioning
US20230306405A1 (en) Payment accessibility notifications
CN117336686B (zh) 一种短信推送方法、装置、电子设备和存储介质
US11862165B1 (en) Optimized virtual assistant for connecting a user to a live agent
US11716295B2 (en) System and method of automated communications via verticalization
US11250955B2 (en) Systems and methods for health evaluation and monitoring
US20230015321A1 (en) System and method for generating a response to a user query

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19896733

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19896733

Country of ref document: EP

Kind code of ref document: A1