CN113095724A - Data processing method and device - Google Patents

Data processing method and device Download PDF

Info

Publication number
CN113095724A
CN113095724A CN202110484445.1A CN202110484445A CN113095724A CN 113095724 A CN113095724 A CN 113095724A CN 202110484445 A CN202110484445 A CN 202110484445A CN 113095724 A CN113095724 A CN 113095724A
Authority
CN
China
Prior art keywords
target
recommendation
semantic
semantic element
client
Prior art date
Legal status (The legal status 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 status listed.)
Granted
Application number
CN202110484445.1A
Other languages
Chinese (zh)
Other versions
CN113095724B (en
Inventor
宋雨
丁锐
万明霞
李敬文
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Bank of China Ltd
Original Assignee
Bank of China Ltd
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 Bank of China Ltd filed Critical Bank of China Ltd
Priority to CN202110484445.1A priority Critical patent/CN113095724B/en
Publication of CN113095724A publication Critical patent/CN113095724A/en
Application granted granted Critical
Publication of CN113095724B publication Critical patent/CN113095724B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • 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
    • G06F16/33—Querying
    • G06F16/335—Filtering based on additional data, e.g. user or group profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/205—Parsing
    • G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/03—Credit; Loans; Processing thereof

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Strategic Management (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Economics (AREA)
  • Development Economics (AREA)
  • Educational Administration (AREA)
  • Accounting & Taxation (AREA)
  • General Business, Economics & Management (AREA)
  • Finance (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Technology Law (AREA)
  • Tourism & Hospitality (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Game Theory and Decision Science (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • Machine Translation (AREA)

Abstract

The invention discloses a data processing method and a data processing device, which can calculate the most urgent structural element required to be clear of a question sentence after obtaining the question sentence input by a client, determine the missing semantic element value of the question sentence by inquiring the client, improve the intention of the client, finally return an accurate answer to the client and improve the accuracy of the answer provided for the problem of the client, thereby improving the effectiveness of answering the client and avoiding the consumption of operation resources and the increase of the work load of electronic equipment caused by the repeated input of the problem by the client.

Description

Data processing method and device
Technical Field
The present invention relates to the field of data processing technologies, and in particular, to a data processing method and apparatus.
Background
With the improvement of intelligent science and technology, the deployment and application range of the bank to the website robot is continuously expanded.
The network robot can replace a teller to communicate with the client by voice to solve the problem proposed by the client. Specifically, the website robot can receive question voice of a client, convert the question voice into a corresponding target question text through voice recognition, then find out a standard question sentence matched with the target question text in a pre-stored standard question sentence set, and then convert an answer text matched with the standard question sentence into a corresponding answer voice and output the answer voice. In the process of searching for the standard question sentences matched with the target question text, one or more standard question sentences can be searched.
However, when finding out multiple standard question sentences, the prior art outputs multiple corresponding answer voices to the client, which may result in that the client cannot effectively obtain the required answers.
Disclosure of Invention
In view of the above problems, the present invention provides a data processing method and apparatus for overcoming the above problems or at least partially solving the above problems, and the technical solution is as follows:
a method of data processing, comprising:
obtaining a target semantic element value in a question sentence input by a client;
finding out at least one target syntactic structure containing target structural elements from a pre-stored syntactic structure set, wherein the syntactic structure set is composed of at least one syntactic structure, each syntactic structure is composed of structural elements of at least one category which are arranged in order, one category of structural elements comprises at least one semantic element category, one semantic element category comprises at least one semantic element value, and the target structural elements are structural elements of the category corresponding to the target semantic element value;
determining at least one piece of recommendation data, wherein the recommendation data comprises a recommendation order of each structural element in a recommendation syntax structure, the number of semantic element classes under each structural element in the recommendation syntax structure, and the number of semantic element values under each structural element in the recommendation syntax structure, and the recommendation syntax structure is one target syntax structure;
respectively inputting the recommended data into an element importance evaluation formula, and respectively obtaining evaluation scores output by the element importance evaluation formula;
determining the evaluation score with the highest score as a target evaluation score among the evaluation scores output by the element importance evaluation formula;
sequentially outputting semantic element values of semantic element categories under corresponding structural elements according to a recommendation sequence in target recommendation data, wherein the target recommendation data are recommendation data corresponding to the target evaluation score;
determining at least one target element value selected by a client in sequence, wherein each target element value is a semantic element value output according to a recommendation order in the target recommendation data;
determining target standard question sentences matched with the target element values from a pre-stored standard question sentence set, wherein the standard question sentence set is composed of at least one standard question sentence;
and outputting the answer matched with the target standard question sentence.
Optionally, the method further includes:
obtaining the question statement input by a customer;
and determining whether the semantics of the question sentence are fuzzy, and if so, executing the step of obtaining the target semantic element value in the question sentence input by the client.
Optionally, the determining whether the semantics of the question statement are fuzzy includes:
determining semantic similarity between the question sentences and each standard question sentence in the standard question sentence set;
and if the semantic similarity between the question statement and at least two standard question statements is not less than a first preset threshold, determining that the semantics of the question statement are fuzzy.
Optionally, the obtaining a target semantic element value in a question sentence input by a client includes:
performing semantic analysis on the question sentences to determine semantic element values in the question sentences;
determining at least one of the semantic element values as a target semantic element value.
Optionally, the element importance evaluation formula is:
Figure BDA0003049755640000031
wherein S is an evaluation score, n is the number of the structural elements, i is the recommended serial number of the structural elements, α i is a depth coefficient, Ci1 is the number of semantic element values under the structural elements corresponding to the serial number i, and Ci2 is the number of semantic element categories under the structural elements corresponding to the serial number i; wherein:
Figure BDA0003049755640000032
optionally, the method further includes:
and correspondingly storing the question sentences and the target standard question sentences.
Optionally, the method further includes:
when it is determined that the client is negatively emotional, the pre-stored lingering conversation is output, and the mode is switched to the manual service mode.
Optionally, the method further includes:
obtaining emotional characteristics of the client, wherein the emotional characteristics comprise at least one of voice characteristics, expression characteristics and limb action characteristics;
comparing the emotion characteristics with emotion similarity of negative emotion characteristics in a pre-stored negative emotion scene;
and when the emotion similarity is not less than a second preset threshold value, determining that the negative emotion appears in the client.
A data processing apparatus comprising: the device comprises a first obtaining unit, a first searching unit, a first determining unit, a first input unit, a second obtaining unit, a second determining unit, a first output unit, a third determining unit, a fourth determining unit and a second output unit, wherein:
the first obtaining unit is configured to perform: obtaining a target semantic element value in a question sentence input by a client;
the first lookup unit configured to perform: finding out at least one target syntactic structure containing target structural elements from a pre-stored syntactic structure set, wherein the syntactic structure set is composed of at least one syntactic structure, each syntactic structure is composed of structural elements of at least one category which are arranged in order, one category of structural elements comprises at least one semantic element category, one semantic element category comprises at least one semantic element value, and the target structural elements are structural elements of the category corresponding to the target semantic element value;
the first determination unit is configured to perform: determining at least one piece of recommendation data, wherein the recommendation data comprises a recommendation order of each structural element in a recommendation syntax structure, the number of semantic element classes under each structural element in the recommendation syntax structure, and the number of semantic element values under each structural element in the recommendation syntax structure, and the recommendation syntax structure is one target syntax structure;
the first input unit configured to perform: respectively inputting each piece of recommended data into an element importance evaluation formula;
the second obtaining unit is configured to perform: respectively obtaining evaluation scores output by the element importance evaluation formula;
the second determination unit configured to perform: determining the evaluation score with the highest score as a target evaluation score among the evaluation scores output by the element importance evaluation formula;
the first output unit configured to perform: sequentially outputting semantic element values of semantic element categories under corresponding structural elements according to a recommendation sequence in target recommendation data, wherein the target recommendation data are recommendation data corresponding to the target evaluation score;
the third determination unit is configured to perform: determining at least one target element value selected by a client in sequence, wherein each target element value is a semantic element value output according to a recommendation order in the target recommendation data;
the fourth determination unit configured to perform: determining target standard question sentences matched with the target element values from a pre-stored standard question sentence set, wherein the standard question sentence set is composed of at least one standard question sentence;
the second output unit configured to perform: and outputting the answer matched with the target standard question sentence.
Optionally, the apparatus further comprises: a third obtaining unit and a fifth determining unit, wherein:
the third obtaining unit is configured to perform: obtaining the question statement input by a customer;
the fifth determination unit configured to perform: and determining whether the semantics of the question sentence are fuzzy, and if so, triggering the first obtaining unit.
Optionally, the fifth determining unit includes: a sixth determining unit and a seventh determining unit, wherein:
the sixth determining unit configured to perform: determining semantic similarity between the question sentences and each standard question sentence in the standard question sentence set;
the seventh determining unit configured to perform: and if the semantic similarity between the question statement and at least two standard question statements is not less than a first preset threshold, determining that the semantics of the question statement are fuzzy, and triggering the first obtaining unit.
Optionally, the first obtaining unit includes: an eighth determining unit and a ninth determining unit, wherein:
the eighth determining unit configured to perform: performing semantic analysis on the question sentences to determine semantic element values in the question sentences;
the ninth determining unit configured to perform: determining at least one of the semantic element values as a target semantic element value.
Optionally, the element importance evaluation formula is:
Figure BDA0003049755640000051
wherein S is an evaluation score, n is the number of the structural elements, i is the recommended serial number of the structural elements, α i is a depth coefficient, Ci1 is the number of semantic element values under the structural elements corresponding to the serial number i, and Ci2 is the number of semantic element categories under the structural elements corresponding to the serial number i; wherein:
Figure BDA0003049755640000052
optionally, the apparatus further comprises: a holding unit;
the saving unit is configured to execute: and correspondingly storing the question sentences and the target standard question sentences.
Optionally, the apparatus further comprises: a third output unit and a switching unit, wherein:
the third output unit configured to perform: upon determining that the client is negatively emotional, outputting a pre-stored leave-on conversation;
the switching unit is configured to perform: and switching to a manual service mode.
Optionally, the apparatus further comprises: a fourth obtaining unit, a comparing unit and a tenth determining unit, wherein:
the fourth obtaining unit is configured to perform: obtaining emotional characteristics of the client, wherein the emotional characteristics comprise at least one of voice characteristics, expression characteristics and limb action characteristics;
the alignment unit is configured to perform: comparing the emotion characteristics with emotion similarity of negative emotion characteristics in a pre-stored negative emotion scene;
the tenth determination unit configured to perform: and when the emotion similarity is not less than a second preset threshold value, determining that the negative emotion appears in the client.
The data processing method and apparatus provided in this embodiment can obtain a target semantic element value in a question sentence input by a client, and find at least one target syntactic structure containing a target structural element from a pre-stored syntactic structure set, where the syntactic structure set is composed of at least one syntactic structure, each syntactic structure is composed of at least one category of structural elements arranged in order, at least one semantic element category is included under one category of structural elements, at least one semantic element value is included under one semantic element category, the target structural element is a structural element of a category corresponding to the target semantic element value, at least one recommendation data is determined, the recommendation data includes a recommendation order of each structural element in the recommendation syntactic structure, a number of semantic element categories under each structural element in the recommendation syntactic structure, and a number of semantic element values under each structural element in the recommendation syntactic structure, the recommendation syntax structure is a target syntax structure, each recommendation data is respectively input into an element importance evaluation formula, evaluation scores output by the element importance evaluation formula are respectively obtained, among the evaluation scores output by the element importance evaluation formula, the evaluation score with the highest score is determined as a target evaluation score, semantic element values of semantic element categories under corresponding structural elements are sequentially output according to the recommendation sequence in the target recommendation data, the target recommendation data is recommendation data corresponding to the target evaluation score, at least one target element value selected by a client in sequence is determined, each target element value is a semantic element value output according to the recommendation sequence in the target recommendation data, target standard problem statements matched with each target element value are determined from a pre-stored standard problem statement set, and the standard problem statement set is composed of at least one standard problem statement, and outputting the answer matched with the target standard question sentence.
According to the invention, after the question sentences input by the client are obtained, the most urgent structural elements required to be definite of the question sentences are calculated, the missing semantic element values of the question sentences are determined by inquiring the client, the intention of the client is perfected, so that accurate answers can be finally returned to the client, and the accuracy of the answers provided for the client questions is improved, thereby improving the effectiveness of answering the client, and avoiding the consumption of operation resources and the increase of work load of electronic equipment caused by repeated input of questions by the client.
The foregoing description is only an overview of the technical solutions of the present invention, and the embodiments of the present invention are described below in order to make the technical means of the present invention more clearly understood and to make the above and other objects, features, and advantages of the present invention more clearly understandable.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, it is obvious that the drawings in the following description are only embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
Fig. 1 is a flowchart illustrating a first data processing method according to an embodiment of the present invention;
FIG. 2 is a flow chart of a second data processing method provided by the embodiment of the invention;
fig. 3 is a schematic structural diagram of a first data processing apparatus according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram illustrating a second data processing apparatus according to an embodiment of the present invention.
Detailed Description
Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.
As shown in fig. 1, the present embodiment proposes a data processing method, which may include the following steps:
s101, obtaining a target semantic element value in a question sentence input by a client;
it should be noted that the present invention can be applied to a website robot, and can also be applied to other electronic devices, such as a mobile phone and a tablet computer.
Wherein the target semantic element value may include at least one semantic element value in the question sentence.
Optionally, step S101 may specifically include:
performing semantic analysis on the question sentences to determine semantic element values in the question sentences;
determining at least one semantic element value as a target semantic element value.
Specifically, after receiving a question sentence input by a client, the method and the device can perform semantic analysis on the question sentence, perform word segmentation on the question sentence, and determine each semantic element value in the question sentence. For example, for a question sentence of "i want to report loss of credit card", the invention can perform semantic analysis on the question sentence, and determine four semantic element values, i.e., "i", "want", "report loss" and "credit card".
Optionally, the target semantic element value can be determined by the semantic element value existing in the preset keyword corpus in the question sentence. For example, in the question sentence "i want to report loss of credit card", the present invention may determine "report loss" and "credit card" existing in a preset keyword corpus as the target semantic element values.
Optionally, the semantic element value belonging to the target part of speech may be determined as the target semantic element value by the present invention. For example, the semantic element value belonging to the verb may be determined as the target semantic element value, the semantic element value belonging to the object may be determined as the target semantic element value, or the semantic element value belonging to the verb and the object may be determined as the target semantic element value.
When a customer inputs a question sentence through voice, the method can acquire corresponding text data through semantic recognition, and then acquire a target semantic element value from the text data.
S102, finding out at least one target syntactic structure containing target structural elements from a pre-stored syntactic structure set, wherein the syntactic structure set is composed of at least one syntactic structure, each syntactic structure is composed of structural elements of at least one category which are arranged in order, the structural elements of one category comprise at least one semantic element category, the semantic element category comprises at least one semantic element value, and the target structural elements are structural elements of the category corresponding to the target semantic element value;
it should be noted that the present invention can obtain the syntactic structure set by performing syntactic analysis on the pre-stored candidate knowledge points.
The candidate knowledge points comprise matched standard question sentences and answer texts.
Specifically, the method can perform syntactic analysis on all standard problem sentences to obtain syntactic dependency trees, and identify the dependency relationships (such as a dominance-predicate relationship, a motile relationship, a core relationship and the like) between words in each standard problem sentence to determine the syntactic structures existing in each standard problem sentence, so as to obtain a syntactic structure set, and determine the semantic element values and the semantic element categories existing under the structural elements of each syntactic structure. The present invention proposes and describes a process of determining semantic element values and semantic element categories existing under each structural element in a syntax structure set and a syntax structure set, in conjunction with example 1.
Example 1: standard question sentences in the candidate knowledge points comprise credit card transaction, credit card loss report, mobile banking credit card loss report, debit card complement, ATM debit card loss report, credit card staging and ice and snow card preference, all the standard question sentences can be analyzed in syntax to determine that three syntax structures exist in each standard question sentence, namely SA + SB + SC (mobile banking/credit card/loss report, ATM/debit card/loss report), SB + SC (credit card/transaction, credit card/loss report, debit card/complement card, credit card/staging) and SB + SD (ice and snow card/benefit), wherein SA, SB, SC and SD can be structural elements in the structure, so that a syntax structure set can be determined; the invention can determine the semantic element value under the structural element, for example, for the structural element SB, the semantic element value comprises a credit card, a debit card, a credit card and an ice and snow card, the number of the semantic element value is 7, the invention can determine the semantic element category under the structural element, for example, for the structural element SB, the semantic element category under the SB comprises two types of credit card and debit card, namely the number of the semantic element category is 2, because the credit card is equal to the credit card and the ice and snow card belongs to one debit card; for another example, for the structural element SA, the semantic element values include mobile banking and ATM, the number of semantic element values is 2, and the number of semantic element categories is 2.
Specifically, the present invention may determine, after the target semantic element value is determined, the structural element corresponding to the target semantic element value as the target structural element, and then may determine, from the set of syntax structures, the syntax structure including the target structural element as the target syntax structure. For example, in the syntax structure set and the structure elements in example 1, when the client input "loss report" is obtained, the present invention may determine the structure element SC corresponding to the "loss report" as the target structure element, and then may determine both SA + SB + SC and SB + SC including the structure element SC as the target syntax structures from the syntax structure set.
Optionally, the invention may compare the semantic element values in all standard question sentences pairwise by using a trained word vector language model, and determine the semantic element values belonging to the same category, so as to determine the semantic element categories and the number of the semantic element categories from the semantic element values under the structural elements.
The semantic element value obtained after syntactic analysis is carried out on the standard question sentence can be utilized to carry out word vector coding on the semantic element value, and a training sample is generated. For example, two semantic element values belonging to the same category are subjected to word vector coding, the coded data "belonging to the same category" is marked, and the marked coded data is determined as a positive sample.
It can be understood that, when the trained word vector language model is used to determine whether two semantic element values belong to the same semantic element category, the word vector language model can compare cosine distances of vectors corresponding to the two semantic element values, and when the cosine distances are smaller than a preset threshold value, it can be determined that the two semantic element values belong to the same semantic element category.
S103, determining at least one piece of recommendation data, wherein the recommendation data comprise a recommendation sequence of each structural element in a recommendation syntax structure, the number of semantic element categories under each structural element in the recommendation syntax structure and the number of semantic element values under each structural element in the recommendation syntax structure, and the recommendation syntax structure is a target syntax structure;
wherein, the recommendation syntax structure may be a target syntax structure.
It should be noted that, the invention can recommend semantic element values possibly missing in the question text to the customer after obtaining the question text input by the customer, obtain the semantic element values selected by the customer, determine the corresponding standard question sentences according to the semantic element values selected by the customer, return answers matched with the standard question sentences to the customer, and improve the accuracy and effectiveness of answer recommendation.
To better illustrate the components in the recommendation data, the present invention is described with reference to example 1 above. Specifically, when the question text input by the client is "loss report", the invention can determine the target syntax structure SA + SB + SC and SB + SC; then, the present invention may use the target syntax structure SA + SB + SC as the recommendation syntax structure, and at this time, SA and SB may be recommended as the structural elements with missing question text, and the recommendation order may be "the recommendation order of SA is 1 and the recommendation order of SB is 2", or "the recommendation order of SB is 1 and the recommendation order of SA is 2", where "loss report" is included in SC, and thus SC may be considered as the structural element recommended by default and selected by the client. Then, the number of semantic element values and the number of semantic element classes under the component can be obtained accordingly, for example, for the component SB, it can be obtained from the above example 1 that the number of semantic element values under SB is 7 and the number of semantic element classes is 2; in the present invention, the target syntax structure SB + SC may be used as the recommendation syntax structure to generate the corresponding recommendation data, and in this case, only SB in the recommendation syntax structure may be recommended as a structural element missing in the question text.
It is understood that the present invention may generate corresponding recommendation data using each of the target syntax structures, respectively, so that a plurality of recommendation data may be obtained.
S104, inputting the recommended data into an element importance evaluation formula;
wherein, the element importance evaluation formula can be used for carrying out quantitative evaluation on the importance scores of the recommendation orders of the structural elements in the recommendation data.
It should be noted that, when the importance score corresponding to the recommendation data is higher, the importance of the recommendation order of the structural elements can be determined to be higher; the present invention can determine that the importance of the recommendation order of the structural elements thereof is lower as the importance score corresponding to the recommendation data is lower.
S105, respectively obtaining evaluation scores output by the factor importance evaluation formula;
wherein the evaluation score can be used to quantitatively evaluate the importance of the recommendation order of the structural elements in the recommendation data.
Alternatively, the element importance evaluation formula may be:
Figure BDA0003049755640000111
wherein S is an evaluation score, n is the number of the structural elements, i is the recommended serial number of the structural elements, α i is a depth coefficient, Ci1 is the number of semantic element values under the structural elements corresponding to the serial number i, and Ci2 is the number of semantic element categories under the structural elements corresponding to the serial number i; wherein:
Figure BDA0003049755640000112
to better illustrate the formula (1) and the formula (2), the present invention is described in conjunction with the above example 1.
Optionally, when the syntax structure of recommendation is SB + SC, only SB can be recommended as a structural element of missing question text, and at this time, the recommendation order of SB is 1 (the recommendation order of SC can be considered as 0), then the corresponding evaluation score of formula (1) is:
Figure BDA0003049755640000113
where i is 1, it is found from equation (2) that α 1 has a value of 0.01, C12, i.e., the number of semantic element classes under SB, is 2, and C11, i.e., the number of semantic element classes under SB, is 7.
Optionally, when the syntax structure of recommendation is SA + SB + SC, there are two recommendation orders, one is "recommendation order of SA is 1, recommendation order of SB is 2", and the other is "recommendation order of SB is 1, recommendation order of SA is 2"; for example, when the recommendation order of SB is 1 and the recommendation order of SA is 2 (the recommendation order of SC can be considered as 0), the corresponding evaluation score of formula (1) is:
Figure BDA0003049755640000114
it is found from the formula (2) that α 1 is 0.01, C12, i.e., the number of semantic element classes under SB is 2, C11, i.e., the number of semantic element classes under SB is 7, α 2 is 0.02, C22, i.e., the number of semantic element classes under SA is 2, and C12, i.e., the number of semantic element classes under SA is 2.
It can be understood that the depth coefficient α i increases as i increases, and as can be seen from formula (1), when the number of structural elements included in the recommendation syntax structure is more, the contribution of the structural element in the following recommendation order to the evaluation score is smaller, and the importance is lower.
S106, determining the evaluation score with the highest score as a target evaluation score in all evaluation scores output by the element importance evaluation formula;
specifically, in each evaluation score output by the element importance evaluation formula for each recommendation datum, the evaluation score with the highest score can be determined as the target evaluation score.
S107, sequentially outputting semantic element values of semantic element categories under corresponding structural elements according to a recommendation sequence in target recommendation data, wherein the target recommendation data are recommendation data corresponding to a target evaluation score;
to better explain the process of sequentially outputting semantic element values according to the recommendation order in the target recommendation data, the present invention is described with reference to example 1. If the invention obtains the loss report input by the client, and the target recommendation data is determined to comprise the recommendation order of each structural element, the number of semantic element categories under each structural element and the number of semantic element values in the recommendation syntax structure SA + SB + SC, and the recommendation order of SA is 1 and the recommendation order of SB is 2, the invention can output the semantic element values of each semantic element category under the structural element SA in advance and then output the semantic element values of each semantic element category under the structural element SB.
Specifically, when the semantic element value of each semantic element category under the SA is output, a "mobile phone bank" and an "ATM" may be output, where the mobile phone bank is the semantic element value under the next semantic element category of the SA, and the ATM is the semantic element value under the other semantic element category of the SA; when outputting semantic element values of semantic element classes under the SB, a credit card/debit card, which is a semantic element value under the same semantic element class under the SB, and an ice card/debit card, which is a semantic element value under the same semantic element class under the SB, may be output.
It should be noted that, after outputting the next semantic element value of a certain structural element, the present invention can wait for the client to select the semantic element value missing from the question sentence.
Optionally, the present invention may output the semantic element value under a certain structural element with a prior recommendation order, and output the semantic element value under the structural element with a subsequent recommendation order after the client selects the semantic element value. For example, when the recommendation order of SA is 1 and the recommendation order of SB is 2, the present invention may output the semantic element value under SA in advance, and output the semantic element value under SB after the client selects the semantic element value under SA.
Optionally, after the semantic element values under the structural elements are output, the client selects the semantic elements missing in the question sentences from the semantic element values. For example, when the recommended order of SA is 1 and the recommended order of SB is 2, the present invention may output the semantic element value under SA first, then output the semantic element value under SB, and then feedback the selected semantic element value under SA and the selected semantic element value under SB by the client. In this case, the present invention does not limit the order of selection of semantic element values by the client for each component.
S108, determining at least one target element value selected by a client in sequence, wherein each target element value is a semantic element value output according to a recommendation sequence in target recommendation data;
specifically, the invention can determine a target element value selected by the customer from each structural element. For example, for the SA and SB, the present invention can determine the "mobile banking" selected by the customer from the SA and the "credit card" selected from the SB.
S109, determining target standard question sentences matched with all target element values from a pre-stored standard question sentence set, wherein the standard question sentence set is composed of at least one standard question sentence;
specifically, the invention can determine the corresponding target standard question sentence according to the question sentence input by the client and the selected target element value. For example, when the question and sentence input by the customer is "loss report" and the selected target element value is "mobile banking" and "credit card", the invention can determine that the target standard question and sentence is "mobile banking loss report credit card".
And S110, outputting the answer matched with the target standard question sentence.
After the target standard question sentence is determined, the answer text matched with the target standard question sentence can be found from the pre-stored candidate knowledge points.
Optionally, the invention can directly output the searched answer text, and also can convert the answer text into corresponding voice data and output the voice data.
It should be noted that, the invention can deduce the most urgent structural elements of the question and sentence needed to be definite after obtaining the question and sentence input by the customer, and determine the missing semantic element value of the question and sentence by asking the customer back, according to the interactive mode of continuous inquiry, gradually improve the intention of the customer, so as to finally return the accurate answer to the customer, improve the accuracy of the answer provided for the customer question, thus improve the effectiveness of answering the customer, and avoid the consumption of the operation resources and the increase of the work load of the electronic equipment caused by the repeated input of the question by the customer.
The data processing method provided in this embodiment may obtain a target semantic element value in a question sentence input by a client, find at least one target syntax structure containing a target structural element from a pre-stored syntax structure set, determine at least one recommendation data, input each recommendation data into an element importance evaluation formula, respectively obtain evaluation scores output by the element importance evaluation formula, determine, as a target evaluation score, an evaluation score with the highest score among the evaluation scores output by the element importance evaluation formula, sequentially output semantic element values of semantic element categories under corresponding structural elements according to a recommendation order in the target recommendation data, determine at least one target element value selected by the client sequentially, determine, from a pre-stored standard question sentence set, a target standard question sentence matching each target element value, and outputting the answer matched with the target standard question sentence. According to the invention, after the question sentences input by the client are obtained, the most urgent structural elements required to be definite of the question sentences are calculated, the missing semantic element values of the question sentences are determined by inquiring the client, the intention of the client is perfected, so that accurate answers can be finally returned to the client, and the accuracy of the answers provided for the client questions is improved, thereby improving the effectiveness of answering the client, and avoiding the consumption of operation resources and the increase of work load of electronic equipment caused by repeated input of questions by the client.
Based on the steps shown in fig. 1, the present embodiment proposes a second data processing method, as shown in fig. 2. The method may further comprise the steps of:
s201, obtaining question sentences input by a client;
the invention can obtain question sentences input by a client in a text or voice mode.
S202, determining whether the semantics of the question sentence are fuzzy, and if so, executing the step S101.
Specifically, the present invention may execute step S101 when the semantic meaning of the question sentence is determined to be fuzzy.
Optionally, step S202 may specifically include:
determining semantic similarity between the question sentences and each standard question sentence in the standard question sentence set;
if the semantic similarity between the question statement and at least two standard question statements is not less than a first preset threshold, determining that the semantics of the question statement are fuzzy, and then executing the step S101.
The specific value of the first preset threshold may be set by a technician according to an actual situation, which is not limited by the present invention.
Specifically, the invention can search the corresponding target standard question sentences from the standard question sentence set in advance after the question sentences are obtained. If more than two target standard sentences are found, the invention can determine the ambiguity of the question sentence and trigger the execution of the step S101.
It should be noted that, when the only target standard question statement matching with the question statement is found out from the standard question statement set, the invention can determine that the question statement is not fuzzy, at this time, the invention can directly return the only answer matching with the target standard question statement to the client, and can prohibit the execution of the step S101, avoid the meaningless consumption of the operation resource,
the data processing method provided by the embodiment can determine whether the problem statement is fuzzy or not after the problem statement is obtained, and only when the semantics are fuzzy, the execution of the step S101 is triggered, and when the semantics are not fuzzy, the execution of the step S101 is forbidden, so that the meaningless consumption of the operation resources is avoided, and the utilization rate of the operation resources is improved.
Based on the steps shown in fig. 1, the present embodiment proposes a third data processing method. The method may further comprise:
s301, storing the question sentences and the target standard question sentences correspondingly.
Specifically, the invention can associate and store the question sentence originally input by the client and the determined target standard question sentence after determining the only target standard question sentence according to the feedback of the client, so that when the problem sentence is encountered again next time, the invention can determine the target standard question sentence matched with the question sentence, output the answer matched with the target standard question sentence, try to solve the problem of the client and effectively save the consumption of the operation resources.
The data processing method provided by the embodiment can correlate and store the question statement originally input by the client and the found target standard question statement, so that when the question statement is encountered again next time, the target standard question statement matched with the question statement can be directly determined, the answer matched with the target standard question statement is output, and the consumption of operation resources is effectively saved.
Based on the steps shown in fig. 1, the present embodiment proposes a fourth data processing method. The method may further comprise:
s401, when determining that the client has negative emotion, outputting a pre-stored lingering conversation, and switching to a manual service mode.
Specifically, when negative emotions such as restlessness and anger are generated, the invention can output a leave-in-speech (for example, the invention can 'do I think again') and switch to a manual service mode, so that the customer is manually served to sooth the emotion of the customer, provide better service for the customer and improve the effectiveness of answering the customer.
It should be noted that, the present invention may also output a leave-talk operation when the recommended semantic element value fed back by the client is not related to its intention, or the recommended semantic element value is wrong information, and switch to a manual service mode, so as to manually service the client.
The invention can judge the emotional characteristics of the client and determine whether the client has negative emotion.
It is understood that the above step S401 can also be applied to the above data processing methods.
Optionally, the method may further include:
obtaining emotional characteristics of the client, wherein the emotional characteristics comprise at least one of voice characteristics, expression characteristics and limb action characteristics;
comparing the emotion characteristics with emotion similarity of negative emotion characteristics in a pre-stored negative emotion scene;
and when the emotion similarity is not less than a second preset threshold value, determining that the negative emotion appears in the client.
Specifically, the invention can obtain the image containing the expression and the limb action of the customer through the camera, and analyze the image to determine the expression characteristic and the limb action characteristic of the customer; speech data containing the voice of the customer is obtained by a voice recorder and analyzed to determine the speech characteristics of the customer.
Then, the emotional characteristics of the client can be compared with the negative emotional characteristics in the typical negative emotional scene, and when the similarity is larger, the negative emotion of the client is determined.
The second preset threshold may be set by a technician according to an actual situation, which is not limited in the present invention.
The data processing method provided by the embodiment can output the leaving dialect and switch to the manual service mode when the negative emotions such as impatience, anger and the like appear in the client, so that the client is manually served, better service is provided for the client, and the effectiveness of answering the client is improved.
Corresponding to the method shown in fig. 1, as shown in fig. 3, the present embodiment proposes a first data processing apparatus, which may include: a first obtaining unit 101, a finding unit 102, a first determining unit 103, a first input unit 104, a second obtaining unit 105, a second determining unit 106, a first output unit 107, a third determining unit 108, a fourth determining unit 109, and a second output unit 110, wherein:
a first obtaining unit 101 configured to perform: obtaining a target semantic element value in a question sentence input by a client;
it should be noted that the present invention can be applied to a website robot, and can also be applied to other electronic devices, such as a mobile phone and a tablet computer.
Wherein the target semantic element value may include at least one semantic element value in the question sentence.
Optionally, the first obtaining unit 101 may include: an eighth determining unit and a ninth determining unit, wherein:
an eighth determination unit configured to perform: performing semantic analysis on the question sentences to determine semantic element values in the question sentences;
a ninth determining unit configured to perform: determining at least one semantic element value as a target semantic element value.
Specifically, after receiving a question sentence input by a client, the method and the device can perform semantic analysis on the question sentence, perform word segmentation on the question sentence, and determine each semantic element value in the question sentence.
Optionally, the target semantic element value can be determined by the semantic element value existing in the preset keyword corpus in the question sentence.
Optionally, the semantic element value belonging to the target part of speech may be determined as the target semantic element value by the present invention.
When a customer inputs a question sentence through voice, the method can acquire corresponding text data through semantic recognition, and then acquire a target semantic element value from the text data.
A lookup unit 102 configured to perform: finding out at least one target syntactic structure containing target structural elements from a pre-stored syntactic structure set, wherein the syntactic structure set consists of at least one syntactic structure, each syntactic structure consists of structural elements of at least one category which are arranged in order, the structural elements of one category comprise at least one semantic element category, the semantic element category comprises at least one semantic element value, and the target structural elements are structural elements of the category corresponding to the target semantic element value;
it should be noted that the present invention can obtain the syntactic structure set by performing syntactic analysis on the pre-stored candidate knowledge points.
The candidate knowledge points comprise matched standard question sentences and answer texts.
Specifically, the method can perform syntactic analysis on all standard problem sentences to obtain syntactic dependency trees, identify the dependency relationship between words in each standard problem sentence to determine the syntactic structures existing in each standard problem sentence, so as to obtain syntactic structure sets, and determine the semantic element values and semantic element categories existing under the structural elements of each syntactic structure.
Specifically, the present invention may determine, after the target semantic element value is determined, the structural element corresponding to the target semantic element value as the target structural element, and then may determine, from the set of syntax structures, the syntax structure including the target structural element as the target syntax structure.
Optionally, the invention may compare the semantic element values in all standard question sentences pairwise by using a trained word vector language model, and determine the semantic element values belonging to the same category, so as to determine the semantic element categories and the number of the semantic element categories from the semantic element values under the structural elements.
The semantic element value obtained after syntactic analysis is carried out on the standard question sentence can be utilized to carry out word vector coding on the semantic element value, and a training sample is generated.
It can be understood that, when the trained word vector language model is used to determine whether two semantic element values belong to the same semantic element category, the word vector language model can compare cosine distances of vectors corresponding to the two semantic element values, and when the cosine distances are smaller than a preset threshold value, it can be determined that the two semantic element values belong to the same semantic element category.
A first determination unit 103 configured to perform: determining at least one piece of recommendation data, wherein the recommendation data comprises a recommendation sequence of each structural element in a recommendation syntax structure, the number of semantic element categories under each structural element in the recommendation syntax structure and the number of semantic element values under each structural element in the recommendation syntax structure, and the recommendation syntax structure is a target syntax structure;
wherein, the recommendation syntax structure may be a target syntax structure.
It should be noted that, the invention can recommend semantic element values possibly missing in the question text to the customer after obtaining the question text input by the customer, obtain the semantic element values selected by the customer, determine the corresponding standard question sentences according to the semantic element values selected by the customer, return answers matched with the standard question sentences to the customer, and improve the accuracy and effectiveness of answer recommendation.
It is understood that the present invention may generate corresponding recommendation data using each of the target syntax structures, respectively, so that a plurality of recommendation data may be obtained.
A first input unit 104 configured to perform: respectively inputting each recommended data into an element importance evaluation formula;
wherein, the element importance evaluation formula can be used for carrying out quantitative evaluation on the importance scores of the recommendation orders of the structural elements in the recommendation data.
It should be noted that, when the importance score corresponding to the recommendation data is higher, the importance of the recommendation order of the structural elements can be determined to be higher; the present invention can determine that the importance of the recommendation order of the structural elements thereof is lower as the importance score corresponding to the recommendation data is lower.
A second obtaining unit 105 configured to perform: respectively obtaining evaluation scores output by the factor importance evaluation formula;
wherein the evaluation score can be used to quantitatively evaluate the importance of the recommendation order of the structural elements in the recommendation data.
Alternatively, the element importance evaluation formula may be:
Figure BDA0003049755640000191
wherein S is an evaluation score, n is the number of the structural elements, i is the recommended serial number of the structural elements, α i is a depth coefficient, Ci1 is the number of semantic element values under the structural elements corresponding to the serial number i, and Ci2 is the number of semantic element categories under the structural elements corresponding to the serial number i; wherein:
Figure BDA0003049755640000192
it can be understood that the depth coefficient α i increases as i increases, and as can be seen from formula (1), when the number of structural elements included in the recommendation syntax structure is more, the contribution of the structural element in the following recommendation order to the evaluation score is smaller, and the importance is lower.
A second determining unit 106 configured to perform: determining the evaluation score with the highest score as a target evaluation score in all evaluation scores output by the element importance evaluation formula;
specifically, in each evaluation score output by the element importance evaluation formula for each recommendation datum, the evaluation score with the highest score can be determined as the target evaluation score.
A first output unit 107 configured to perform: sequentially outputting semantic element values of semantic element categories under corresponding structural elements according to a recommendation sequence in the target recommendation data, wherein the target recommendation data are recommendation data corresponding to the target evaluation score;
it should be noted that, after outputting the next semantic element value of a certain structural element, the present invention can wait for the client to select the semantic element value missing from the question sentence.
Optionally, the present invention may output the semantic element value under a certain structural element with a prior recommendation order, and output the semantic element value under the structural element with a subsequent recommendation order after the client selects the semantic element value.
Optionally, after the semantic element values under the structural elements are output, the client selects the semantic elements missing in the question sentences from the semantic element values.
A third determining unit 108 configured to perform: determining at least one target element value selected by a client in sequence, wherein each target element value is a semantic element value output according to a recommendation order in target recommendation data;
specifically, the invention can determine a target element value selected by the customer from each structural element.
A fourth determination unit 109 configured to perform: determining target standard question sentences matched with all target element values from a pre-stored standard question sentence set, wherein the standard question sentence set is composed of at least one standard question sentence;
specifically, the invention can determine the corresponding target standard question sentence according to the question sentence input by the client and the selected target element value.
A second output unit 110 configured to perform: and outputting the answer matched with the target standard question sentence.
After the target standard question sentence is determined, the answer text matched with the target standard question sentence can be found from the pre-stored candidate knowledge points.
Optionally, the invention can directly output the searched answer text, and also can convert the answer text into corresponding voice data and output the voice data.
It should be noted that, the invention can deduce the most urgent structural elements of the question and sentence needed to be definite after obtaining the question and sentence input by the customer, and determine the missing semantic element value of the question and sentence by asking the customer back, according to the interactive mode of continuous inquiry, gradually improve the intention of the customer, so as to finally return the accurate answer to the customer, improve the accuracy of the answer provided for the customer question, thus improve the effectiveness of answering the customer, and avoid the consumption of the operation resources and the increase of the work load of the electronic equipment caused by the repeated input of the question by the customer.
The data processing device provided by the embodiment can deduce the most urgent structural elements required to be clear of the question sentences after obtaining the question sentences input by the client, determine the missing semantic element values of the question sentences by inquiring the client, perfect the intention of the client, finally return accurate answers to the client, and improve the accuracy of the answers provided for the client, so that the effectiveness of answering the client is improved, and the consumption of operation resources and the increase of work load of electronic equipment caused by repeated input of questions by the client are avoided.
Based on the schematic structural diagram shown in fig. 3, as shown in fig. 4, the present embodiment proposes a second data processing apparatus, which may further include: a third obtaining unit 201 and a fifth determining unit 202, wherein:
a third obtaining unit 201 configured to perform: obtaining a question statement input by a customer;
the invention can obtain question sentences input by a client in a text or voice mode.
A fifth determining unit 202 configured to perform: it is determined whether the semantics of the question statement are ambiguous and if so, the first obtaining unit 101 is triggered.
Specifically, the present invention may trigger the first obtaining unit 101 when the semantic meaning of the question statement is determined to be fuzzy.
Optionally, the fifth determining unit 202 includes: a sixth determining unit and a seventh determining unit, wherein:
a sixth determination unit configured to perform: determining semantic similarity between the question sentences and each standard question sentence in the standard question sentence set;
a seventh determining unit configured to perform: if the semantic similarity between the question statement and at least two standard question statements is not less than a first preset threshold, determining that the semantics of the question statement are fuzzy, and triggering the first obtaining unit 101.
The specific value of the first preset threshold may be set by a technician according to an actual situation, which is not limited by the present invention.
Specifically, the invention can search the corresponding target standard question sentences from the standard question sentence set in advance after the question sentences are obtained. If more than two target standard sentences are found, the present invention can determine the ambiguity of the question sentence and can trigger the first obtaining unit 101.
It should be noted that, when the only target standard question statement matched with the question statement is found out from the standard question statement set, the invention can determine that the question statement is not fuzzy, at this time, the invention can directly return the only answer matched with the target standard question statement to the client, and can forbid triggering the first obtaining unit 101, thereby avoiding meaningless consumption of operation resources,
the data processing apparatus provided in this embodiment may determine whether the problem statement is ambiguous or not after the problem statement is obtained, trigger the first obtaining unit 101 only when the semantic is ambiguous, and prohibit triggering the first obtaining unit 101 when the semantic is not ambiguous, thereby avoiding unnecessary consumption of the operation resources and improving the utilization rate of the operation resources.
Based on the schematic structural diagram shown in fig. 3, the present embodiment provides a third data processing apparatus, which may further include: a holding unit;
a saving unit configured to perform: and correspondingly storing the question sentences and the target standard question sentences.
Specifically, the invention can associate and store the question sentence originally input by the client and the determined target standard question sentence after determining the only target standard question sentence according to the feedback of the client, so that when the problem sentence is encountered again next time, the invention can determine the target standard question sentence matched with the question sentence, output the answer matched with the target standard question sentence, try to solve the problem of the client and effectively save the consumption of the operation resources.
The data processing device provided by the embodiment can correlate and store the question statement originally input by the client and the searched target standard question statement, so that when the question statement is encountered again next time, the target standard question statement matched with the question statement can be directly determined, the answer matched with the target standard question statement is output, and the consumption of operation resources is effectively saved.
Based on the schematic structure diagram shown in fig. 3, the present embodiment provides a fourth data processing apparatus. The apparatus may further include: a third output unit and a switching unit, wherein:
a third output unit configured to perform: upon determining that the client is negatively emotional, outputting a pre-stored leave-on conversation;
a switching unit configured to perform: and switching to a manual service mode.
Specifically, when negative emotions such as restlessness and anger are generated, the invention can output a leave-in-speech (for example, the invention can 'do I think again') and switch to a manual service mode, so that the customer is manually served to sooth the emotion of the customer, provide better service for the customer and improve the effectiveness of answering the customer.
It should be noted that, the present invention may also output a leave-talk operation when the recommended semantic element value fed back by the client is not related to its intention, or the recommended semantic element value is wrong information, and switch to a manual service mode, so as to manually service the client.
The invention can judge the emotional characteristics of the client and determine whether the client has negative emotion.
It is to be understood that the third output unit and the switching unit may also be applied to the data processing apparatuses.
Optionally, the apparatus may further include: a fourth obtaining unit, a comparing unit and a tenth determining unit, wherein:
a fourth obtaining unit configured to perform: obtaining emotional characteristics of the client, wherein the emotional characteristics comprise at least one of voice characteristics, expression characteristics and limb action characteristics;
an alignment unit configured to perform: comparing the emotion characteristics with emotion similarity of negative emotion characteristics in a pre-stored negative emotion scene;
a tenth determination unit configured to perform: and when the emotion similarity is not less than a second preset threshold value, determining that the negative emotion appears in the client.
Specifically, the invention can obtain the image containing the expression and the limb action of the customer through the camera, and analyze the image to determine the expression characteristic and the limb action characteristic of the customer; speech data containing the voice of the customer is obtained by a voice recorder and analyzed to determine the speech characteristics of the customer. Then, the emotional characteristics of the client can be compared with the negative emotional characteristics in the typical negative emotional scene, and when the similarity is larger, the negative emotion of the client is determined.
The second preset threshold may be set by a technician according to an actual situation, which is not limited in the present invention.
The data processing device provided by the embodiment can output the leaving-talk when the negative emotions such as impatience and anger occur to the client, and switch to the manual service mode, so that the client is manually served, better service is provided for the client, and the effectiveness of answering the client is improved.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in the process, method, article, or apparatus that comprises the element.
The above are merely examples of the present application and are not intended to limit the present application. Various modifications and changes may occur to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims (10)

1. A data processing method, comprising:
obtaining a target semantic element value in a question sentence input by a client;
finding out at least one target syntactic structure containing target structural elements from a pre-stored syntactic structure set, wherein the syntactic structure set is composed of at least one syntactic structure, each syntactic structure is composed of structural elements of at least one category which are arranged in order, one category of structural elements comprises at least one semantic element category, one semantic element category comprises at least one semantic element value, and the target structural elements are structural elements of the category corresponding to the target semantic element value;
determining at least one piece of recommendation data, wherein the recommendation data comprises a recommendation order of each structural element in a recommendation syntax structure, the number of semantic element classes under each structural element in the recommendation syntax structure, and the number of semantic element values under each structural element in the recommendation syntax structure, and the recommendation syntax structure is one target syntax structure;
respectively inputting the recommended data into an element importance evaluation formula, and respectively obtaining evaluation scores output by the element importance evaluation formula;
determining the evaluation score with the highest score as a target evaluation score among the evaluation scores output by the element importance evaluation formula;
sequentially outputting semantic element values of semantic element categories under corresponding structural elements according to a recommendation sequence in target recommendation data, wherein the target recommendation data are recommendation data corresponding to the target evaluation score;
determining at least one target element value selected by a client in sequence, wherein each target element value is a semantic element value output according to a recommendation order in the target recommendation data;
determining target standard question sentences matched with the target element values from a pre-stored standard question sentence set, wherein the standard question sentence set is composed of at least one standard question sentence;
and outputting the answer matched with the target standard question sentence.
2. The method of claim 1, further comprising:
obtaining the question statement input by a customer;
and determining whether the semantics of the question sentence are fuzzy, and if so, executing the step of obtaining the target semantic element value in the question sentence input by the client.
3. The method of claim 2, wherein the determining whether the semantics of the question statement are ambiguous comprises:
determining semantic similarity between the question sentences and each standard question sentence in the standard question sentence set;
and if the semantic similarity between the question statement and at least two standard question statements is not less than a first preset threshold, determining that the semantics of the question statement are fuzzy.
4. The method of claim 1, wherein obtaining the target semantic element value in the question statement input by the customer comprises:
performing semantic analysis on the question sentences to determine semantic element values in the question sentences;
determining at least one of the semantic element values as a target semantic element value.
5. The method of claim 1, wherein the element importance assessment formula is:
Figure FDA0003049755630000021
wherein S is an evaluation score, n is the number of the structural elements, i is the recommended serial number of the structural elements, α i is a depth coefficient, Ci1 is the number of semantic element values under the structural elements corresponding to the serial number i, and Ci2 is the number of semantic element categories under the structural elements corresponding to the serial number i; wherein:
Figure FDA0003049755630000022
6. the method of claim 1, further comprising:
and correspondingly storing the question sentences and the target standard question sentences.
7. The method of any of claims 1 to 6, further comprising:
when it is determined that the client is negatively emotional, the pre-stored lingering conversation is output, and the mode is switched to the manual service mode.
8. The method of claim 7, further comprising:
obtaining emotional characteristics of the client, wherein the emotional characteristics comprise at least one of voice characteristics, expression characteristics and limb action characteristics;
comparing the emotion characteristics with emotion similarity of negative emotion characteristics in a pre-stored negative emotion scene;
and when the emotion similarity is not less than a second preset threshold value, determining that the negative emotion appears in the client.
9. A data processing apparatus, comprising: the device comprises a first obtaining unit, a searching unit, a first determining unit, a first input unit, a second obtaining unit, a second determining unit, a first output unit, a third determining unit, a fourth determining unit and a second output unit, wherein:
the first obtaining unit is configured to perform: obtaining a target semantic element value in a question sentence input by a client;
the lookup unit configured to perform: finding out at least one target syntactic structure containing target structural elements from a pre-stored syntactic structure set, wherein the syntactic structure set is composed of at least one syntactic structure, each syntactic structure is composed of structural elements of at least one category which are arranged in order, one category of structural elements comprises at least one semantic element category, one semantic element category comprises at least one semantic element value, and the target structural elements are structural elements of the category corresponding to the target semantic element value;
the first determination unit is configured to perform: determining at least one piece of recommendation data, wherein the recommendation data comprises a recommendation order of each structural element in a recommendation syntax structure, the number of semantic element classes under each structural element in the recommendation syntax structure, and the number of semantic element values under each structural element in the recommendation syntax structure, and the recommendation syntax structure is one target syntax structure;
the first input unit configured to perform: respectively inputting each piece of recommended data into an element importance evaluation formula;
the second obtaining unit is configured to perform: respectively obtaining evaluation scores output by the element importance evaluation formula;
the second determination unit configured to perform: determining the evaluation score with the highest score as a target evaluation score among the evaluation scores output by the element importance evaluation formula;
the first output unit configured to perform: sequentially outputting semantic element values of semantic element categories under corresponding structural elements according to a recommendation sequence in target recommendation data, wherein the target recommendation data are recommendation data corresponding to the target evaluation score;
the third determination unit is configured to perform: determining at least one target element value selected by a client in sequence, wherein each target element value is a semantic element value output according to a recommendation order in the target recommendation data;
the fourth determination unit configured to perform: determining target standard question sentences matched with the target element values from a pre-stored standard question sentence set, wherein the standard question sentence set is composed of at least one standard question sentence;
the second output unit configured to perform: and outputting the answer matched with the target standard question sentence.
10. The apparatus of claim 9, further comprising: a third obtaining unit and a fifth determining unit, wherein:
the third obtaining unit is configured to perform: obtaining the question statement input by a customer;
the fifth determination unit configured to perform: and determining whether the semantics of the question sentence are fuzzy, and if so, triggering the first obtaining unit.
CN202110484445.1A 2021-04-30 2021-04-30 Data processing method and device Active CN113095724B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110484445.1A CN113095724B (en) 2021-04-30 2021-04-30 Data processing method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110484445.1A CN113095724B (en) 2021-04-30 2021-04-30 Data processing method and device

Publications (2)

Publication Number Publication Date
CN113095724A true CN113095724A (en) 2021-07-09
CN113095724B CN113095724B (en) 2024-03-01

Family

ID=76681151

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110484445.1A Active CN113095724B (en) 2021-04-30 2021-04-30 Data processing method and device

Country Status (1)

Country Link
CN (1) CN113095724B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102047385B1 (en) * 2018-09-21 2019-11-21 삼성생명보험주식회사 Method, apparatus and computer-readable medium for operating chatbot
CN112069298A (en) * 2020-07-31 2020-12-11 杭州远传新业科技有限公司 Human-computer interaction method, device and medium based on semantic web and intention recognition
WO2020256204A1 (en) * 2019-06-17 2020-12-24 주식회사 오니온파이브 System and method for recommending answer on basis of contents and emotion analysis of text
CN112632248A (en) * 2020-12-22 2021-04-09 深圳追一科技有限公司 Question answering method, device, computer equipment and storage medium

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102047385B1 (en) * 2018-09-21 2019-11-21 삼성생명보험주식회사 Method, apparatus and computer-readable medium for operating chatbot
WO2020256204A1 (en) * 2019-06-17 2020-12-24 주식회사 오니온파이브 System and method for recommending answer on basis of contents and emotion analysis of text
CN112069298A (en) * 2020-07-31 2020-12-11 杭州远传新业科技有限公司 Human-computer interaction method, device and medium based on semantic web and intention recognition
CN112632248A (en) * 2020-12-22 2021-04-09 深圳追一科技有限公司 Question answering method, device, computer equipment and storage medium

Also Published As

Publication number Publication date
CN113095724B (en) 2024-03-01

Similar Documents

Publication Publication Date Title
CN108363790B (en) Method, device, equipment and storage medium for evaluating comments
CN108829893B (en) Method and device for determining video label, storage medium and terminal equipment
CN107330120B (en) Inquire answer method, inquiry answering device and computer readable storage medium
CN107480143B (en) Method and system for segmenting conversation topics based on context correlation
US9740677B2 (en) Methods and systems for analyzing communication situation based on dialogue act information
US6173261B1 (en) Grammar fragment acquisition using syntactic and semantic clustering
US9792279B2 (en) Methods and systems for analyzing communication situation based on emotion information
KR102288249B1 (en) Information processing method, terminal, and computer storage medium
US8204751B1 (en) Relevance recognition for a human machine dialog system contextual question answering based on a normalization of the length of the user input
US20180082184A1 (en) Context-aware chatbot system and method
CA3176868A1 (en) Intent identifying method and device for application to intelligent customer service robot
KR100818979B1 (en) Dialog management apparatus and method for chatting agent
CN112069298A (en) Human-computer interaction method, device and medium based on semantic web and intention recognition
CN110472224B (en) Quality of service detection method, apparatus, computer device and storage medium
CN109815491B (en) Answer scoring method, device, computer equipment and storage medium
CN114580382A (en) Text error correction method and device
CN105912629A (en) Intelligent question and answer method and device
US20230069935A1 (en) Dialog system answering method based on sentence paraphrase recognition
US11163961B2 (en) Detection of relational language in human-computer conversation
CN112818109B (en) Intelligent reply method, medium, device and computing equipment for mail
CN112307164A (en) Information recommendation method and device, computer equipment and storage medium
EP4047506A1 (en) Method and system to automatically train a chatbot using domain conversations
CN112632248A (en) Question answering method, device, computer equipment and storage medium
CN113220854A (en) Intelligent dialogue method and device for machine reading understanding
CN116501858B (en) Text processing and data query method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant