Outbound call scene data query method, device, computer equipment and storage medium

Abstract

The embodiment of the present invention discloses an outbound call scenario data query method, device, computer equipment and storage medium. The method includes: constructing real-time query data for various outbound call scenarios to obtain personalized data; obtaining conversation data for outbound call scenarios; performing question discrimination processing on the conversation data to obtain personalized query sentences; parsing the personalized query sentences into query SQL statements; querying corresponding content from the personalized data according to the query SQL statements to form query results; judging whether the query result is a no-data response result; if not, encapsulating the data corresponding to the query result and outputting the encapsulated data. By implementing the method of the embodiment of the present invention, the outbound call system can be enabled to perform real-time data query capabilities and provide accurate response results for personalized data query needs.

Description

Technical Field

[0001] The present invention relates to computers, and more specifically to an outbound call scene data query method, device, computer equipment and storage medium. Background Art

[0002] In banking and financial intelligent outbound telephone marketing scenarios, such as credit card promotion or credit limit increase, repayment reminders and many other outbound call scenarios, users' personalized data query needs are often encountered, such as "Help me check last month's bill?" or "How much is my current balance?" or "Check how many times I have been overdue in the past two years." For manual customer service, they can generally give correct feedback based on user questions and needs. However, most of the current outbound telephone systems are difficult to correctly solve or guide when faced with word class problems, which makes it impossible to meet users' information consultation expectations, greatly reducing the overall outbound call effect.

[0003] Therefore, it is necessary to design a new method to enable the outbound call system to perform real-time data query capabilities and provide accurate response results for personalized data query needs. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the prior art and provide a method, device, computer equipment and storage medium for querying outbound call scene data.

[0005] To achieve the above object, the present invention adopts the following technical solution: an outbound call scene data query method, comprising:

[0006] Build real-time query data for various outbound call scenarios to obtain personalized data;

[0007] Get the conversation data of the outbound call scenario;

[0008] Performing question sentence discrimination processing on the dialogue data to obtain personalized query sentences;

[0009] Parsing the personalized query into a query SQL statement;

[0010] Querying corresponding content from the personalized data according to the query SQL statement to form a query result;

[0011] Determine whether the query result is a no-data response result;

[0012] If the query result is not a no-data response result, the data corresponding to the query result is encapsulated and the encapsulated data is output.

[0013] A further technical solution is: constructing real-time query data for various outbound call scenarios to obtain personalized data, including:

[0014] Record historical outbound call information according to business scenarios;

[0015] Identify the personalized query data item of each conversation in the historical outbound call information according to the set rules;

[0016] The personalized query data items of each conversation are combined and counted to obtain statistical results;

[0017] The statistical results are used to construct relevant database tables according to common real-time query data items in the set scenarios to obtain personalized data.

[0018] A further technical solution is: the question identification processing of the dialogue data to obtain a personalized query sentence includes:

[0019] Performing word segmentation and entity recognition on the conversation data to obtain entity words;

[0020] Performing entity linking processing on the entity words to obtain semantic information;

[0021] The question type is identified on the semantic information to obtain a personalized query sentence.

[0022] A further technical solution is: performing entity linking processing on the entity words to obtain semantic information includes:

[0023] The entity words are aligned with standardized semantic entities to obtain semantic information.

[0024] A further technical solution is: the identification of the question type of the semantic information to obtain a personalized query sentence includes:

[0025] The semantic information is identified by using a rule and sequence labeling method or a TextCNN short text classification model to obtain a personalized query sentence.

[0026] A further technical solution is: parsing the personalized query sentence into a query SQL statement includes:

[0027] Classify and identify the database table of personalized data by using a similarity calculation method according to the personalized query sentence, so as to determine the database table corresponding to the personalized query sentence;

[0028] Extracting relevant fields from a database table corresponding to the personalized query sentence according to the personalized query sentence to obtain query semantics;

[0029] Identifying semantic incidental information in the personalized query sentence to obtain query conditions;

[0030] The database table corresponding to the personalized query sentence, the query semantics and the query condition are assembled to obtain a query SQL statement.

[0031] A further technical solution is: encapsulating the data corresponding to the query result and outputting the encapsulated data, including:

[0032] The data corresponding to the query result is converted into text through NLG technology, and the text is output through TTS speech synthesis technology.

[0033] The present invention also provides an outbound call scene data query device, comprising:

[0034] Data construction unit, used to construct real-time query data for various outbound call scenarios to obtain personalized data;

[0035] A conversation data acquisition unit, used to acquire conversation data of an outbound call scenario;

[0036] A processing unit, configured to perform question identification processing on the dialogue data to obtain a personalized query sentence;

[0037] A parsing unit, used to parse the personalized query sentence into a query SQL statement;

[0038] A query unit, used to query corresponding content from the personalized data according to the query SQL statement to form a query result;

[0039] A judging unit, used to judge whether the query result is a no-data response result;

[0040] The encapsulation unit is used to encapsulate the data corresponding to the query result if the query result is not a no-data response result, and output the encapsulated data.

[0041] The present invention further provides a computer device, wherein the computer device comprises a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above method when executing the computer program.

[0042] The present invention also provides a storage medium, wherein the storage medium stores a computer program, and the computer program can implement the above method when executed by a processor.

[0043] The beneficial effects of the present invention compared with the prior art are: the present invention constructs personalized data from historical outbound call scenarios, and when a conversation is conducted, question identification, question analysis and query are performed on the conversation data, and the query results are packaged and output, thereby realizing the outbound call system's ability to perform real-time data query capabilities and providing accurate response results for personalized data query needs.

[0044] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0046] Figure 1 A schematic diagram of an application scenario of the outbound call scenario data query method provided by an embodiment of the present invention;

[0047] Figure 2 A flow chart of an outbound call scenario data query method provided by an embodiment of the present invention;

[0048] Figure 3 A schematic diagram of a sub-process of an outbound call scenario data query method provided by an embodiment of the present invention;

[0049] Figure 4 A schematic diagram of a sub-process of an outbound call scenario data query method provided by an embodiment of the present invention;

[0050] Figure 5 A schematic diagram of a sub-process of an outbound call scenario data query method provided by an embodiment of the present invention;

[0051] Figure 6 A schematic diagram of the structure of the Encoder-Decoder model provided in an embodiment of the present invention;

[0052] Figure 7 A schematic block diagram of an outbound call scenario data query device provided by an embodiment of the present invention;

[0053] Figure 8 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0056] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0057] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0058] See also Figure 1 and Figure 2 , Figure 1 A schematic diagram of an application scenario of the outbound call scenario data query method provided in an embodiment of the present invention. Figure 2 A schematic flow chart of an outbound call scenario data query method provided by an embodiment of the present invention. The outbound call scenario data query method is applied in a server. The server interacts with a terminal, obtains conversation data of an outbound call scenario from the terminal, performs question discrimination, analysis and query, and converts the result into voice output to the terminal.

[0059] Figure 2 FIG. 1 is a flow chart of a method for querying outbound call scenario data provided by an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S180.

[0060] S110. Construct real-time query data for various outbound call scenarios to obtain personalized data.

[0061] In this embodiment, the personalized data refers to a plurality of database tables related to personalized questions.

[0062] In one embodiment, see Figure 3 , the above-mentioned step S110 may include steps S111 to S114.

[0063] S111. Record historical outbound call information according to business scenarios.

[0064] In this embodiment, the historical outbound call information refers to the information of the outbound call conversations recorded in the bank outbound call scenario.

[0065] Because outbound marketing is conducted according to business scenarios, for example, credit card promotion and large-denomination certificate of deposit marketing are separate businesses. Therefore, for historical intelligent marketing outbound calls, telephone communication records are made according to business scenarios, and the scenario name, user ID, conversation time, speech content and other information of each conversation are recorded in detail; then, historical outbound call data and historical manual outbound call data are pulled, and further analyzed to obtain data items for personalized real-time queries of users in each outbound call scenario.

[0066] S112: Identify the personalized query data item of each conversation in the historical outbound call information according to the set rules.

[0067] In this embodiment, the personalized query data items are personal data information fields that users often ask about in telephone scenarios and have been prepared in advance in the database table.

[0068] Specifically, the personalized data query items are identified or extracted for each historical outbound call conversation record in the business scenario by using rules or keyword extraction. The specific methods are as follows: The first is to use the rule recognition method to write regular sentence rules for pattern matching. According to the user's personalized query words, such as: "Help me see how much I owe in xx month?", "How many times have I been overdue?", etc., we can write the following sentence rules for personalized data item identification: .*(query|view|look).*(my|me|recent|${time})(.*)(how much|how many times|several|); .*(my|me)(.*)(recent|total|total)(.*)(how much|how many times|several); The second method uses the model extraction and recognition method, using the CRF sequence labeling model to identify the personalized data items in the question. First, some data are manually labeled and annotated by the "BIO" sequence labeling system. The specific label list is shown in Table 1.

[0069] Table 1. Tag list

[0070]

[0071] An example of question annotation is shown in Table 2. For example, the question is: “How many times have I been overdue in the past year?”

[0072] Table 2. Annotation results

[0073] Word number Character Sequence Tags 1 I O 2 close B-Time 3 1 B-Time 4 Year B-Time 5 common B-Data 6 More than I-Data 7 Expect I-Data 8 It's I-Data 9 many I-Data 10 few I-Data 11 Second-rate I-Data 12 ？ O

[0074] For non-personalized data queries, such as "Who are you?", "I'm driving now and I'm busy", etc., the labeling results are all "O". After labeling, the BiLSTM+CRF sequence standard model based on the deep model is used to train the personalized data recognition model in the outbound call scenario. The entire model structure includes the input layer, word embedding vector layer, Bi-LSTM layer and output layer connected in sequence.

[0075] S113: The personalized query data items of each conversation are merged and counted to obtain statistical results.

[0076] In this embodiment, the statistical results refer to personalized query data items classified according to the same query question.

[0077] The personalized data items in each conversation are obtained. However, each user may have different statements about the same data item. For example, "Number of overdue payments", "How many times in total have you overdue", and "How many times have you overdue" all mean to query "Number of overdue payments". Therefore, we merge and count personalized data items according to the scenario. The specific method is: use open source synonym dictionaries such as Xinhua Dictionary, Synonym Forest, etc., and manually compiled synonym dictionaries to merge data items with the same semantics; use similarity calculation methods, such as cosine similarity calculation based on word TFIDF, to merge similar data items together. After the above merging process and manual business sorting, all personalized data items for each scenario are obtained. For example, for the "credit card scenario", personalized data items such as "credit card limit", "Number of overdue personal credit card payments", "Total amount owed", "Repayment date", and "Instalment interest rate" can be obtained, that is, users may ask about these data in the credit card outbound call scenario.

[0078] S114: construct relevant database tables for the statistical results according to the common real-time query data items in the set scenarios to obtain personalized data.

[0079] Due to the wide variety of banking services, there are usually tens of thousands of database tables, and there are data sensitivity and security issues. In addition, because voice outbound calls have high requirements for real-time query response performance of data, it is impossible to directly provide users with database table queries based on the original full data. It is necessary to independently build relevant database tables for real-time query data according to business scenarios based on the common real-time query data items of each scenario obtained through statistics and collation in the above steps. Relational databases such as MYSQL are generally used for data storage. Generally speaking, in a scenario, users will ask about multiple types of data in related aspects, so it is generally necessary to build multiple database tables. The construction rules of the database tables are mainly classified according to the business involved in the user's query, for example: credit card business, including credit card overdue information table, credit card consumption information table, credit card basic limit information table, etc.

[0080] Such as the following database table:

[0081] Table 1 - User basic information table - ID, user code, institution number, institution name, account manager, account opening time, last N digits of ID card...

[0082] Table 2 - User credit card overdue table - ID, user code, overdue amount, overdue time, overdue days...

[0083] Table 3 - User consumption record table - ID, user code, consumption amount, consumption time, consumption channel...

[0084] S120: Obtain conversation data of an outbound call scenario.

[0085] In this embodiment, the conversation data refers to the data recorded during the outbound call process.

[0086] S130: performing question sentence identification processing on the dialogue data to obtain personalized query sentences.

[0087] In this embodiment, the personalized query sentence refers to a sentence that meets the personalized data query category.

[0088] During the outbound call process, it is necessary to accurately identify the questions of these personalized data queries, convert the relevant questions into accurate structured query statements, and then query and retrieve from the database that has been prepared in advance. In the outbound call scenario, users conduct real-time queries on personalized data. The focus of the question-answering understanding stage is to convert the voice text in the outbound call into structured queries. Since TTS (Text To Speech) speech synthesis, ASR (Automatic Speech Recognition), and non-real-time query parsing such as FAQ similar sentence matching and multiple question-answering branches are common parts of general outbound call dialogue systems, this article will not go into details due to the relatively common solutions, but will focus on the real-time query understanding of questions and the encapsulation of query results. In the outbound call system, the user's speech is converted into text-based user questions after ASR speech recognition.

[0089] In one embodiment, see Figure 4 , the above-mentioned step S130 may include steps S131 to S133.

[0090] S131. Perform word segmentation and entity recognition on the conversation data to obtain entity words.

[0091] In this embodiment, the entity words refer to entity words in the conversation data.

[0092] The conversation data is segmented and entity recognized. The specific method is generally to use dictionary matching or deep learning sequence labeling model BiLSTM+CRF, or a combination of dictionary matching and machine learning. Word segmentation or entity recognition is currently a very mature processing technology in natural language processing.

[0093] Since users need to search for certain types of database tables and specific fields when they conduct real-time data queries, these potential semantic information needs to be identified in the entity recognition stage. For example, in the credit repayment reminder business scenario, users often ask "How much money am I overdue this month?" We need to identify the association between "overdue" and "how much money" and the database table. Therefore, the database table for scenario query needs to generate a semantic dictionary. Generally, three business data dictionaries need to be generated: a dictionary of library table names, a dictionary of field names, and a dictionary of library table field content. The dictionary is organized by word type, word classification, synonyms, and other information, as shown in Tables 3 to 5 respectively.

[0094] Table 3. Library table name dictionary

[0095]

[0096] Table 4. Table field name information dictionary

[0097]

[0098]

[0099] Table 5. Library table field contents

[0100]

[0101] In addition to the business semantic entities in the database table, it is also necessary to increase the recognition of general semantic entities such as quantifiers and time to support queries with general numbers, time and other restrictive conditions.

[0102] S132: Perform entity linking processing on the entity words to obtain semantic information.

[0103] In this embodiment, the semantic information refers to the standardized semantic content corresponding to the entity words.

[0104] Specifically, the entity words are aligned with standardized semantic entities to obtain semantic information.

[0105] Entity linking processing refers to aligning the identified entity words with standardized semantic entities to identify more standardized semantic information. Generally speaking, since the words in the semantic dictionary can be searched in a regular manner, the words in the semantic dictionary are relatively simple. However, words such as numbers and time require special calculations to obtain general standardized representation information. For example, in "How many times have I been overdue this year", this year is a time entity, and it is necessary to further identify that its start time is: 2021-01-01, the end time is: 2021-12-31, the corresponding year is: 2021, and the time unit granularity is: year and other semantic information.

[0106] S133: Identify the question type of the semantic information to obtain a personalized query sentence.

[0107] Specifically, the semantic information is identified by using a rule and sequence labeling method or a TextCNN short text classification model to obtain a personalized query sentence.

[0108] In the outbound call scenario, the query of user personalized data is interspersed in the conversation. Therefore, it is necessary to monitor the real user personalized query sentences in real time. The real-time query data discrimination module, like the personalized query data items of each conversation in the outbound call recognition scenario 2, uses rules and sequence annotation methods to identify. When the regular rules are met or the sequence annotation recognition results contain the data query item, it is considered to be a real-time data query sentence. Of course, since this is a binary classification problem, we can also use a classification model to process it. Specifically, we use the TextCNN short text classification model. The entire forward classification logic of the TextCNN short text classification model is: vectorize the representation by word to obtain a two-dimensional vector of [Sen_Length, Embedding_Dim], where Sen_Length is the sentence length and Embedding_Dim is the vectorized representation of each word segment; use N convolution kernels to obtain N convolution feature maps; perform max-pooling on the convolution feature map to obtain N 1x1 values, so that sentences of different lengths can be converted into fixed-length representations after the pooling layer, and then spliced ​​into an N-dimensional vector as the sentence representation of the text. Finally, a fully connected softmax layer is connected to compress the N-dimensional vector to a 1-dimensional vector of length 2, and output the probability of each category. [1, 0] can be used to represent the personalized data query class, and [0, 1] can be used to represent the non-personalized query class.

[0109] S140: Parse the personalized query sentence into a query SQL statement.

[0110] In this embodiment, the query SQL statement refers to a structured query statement, including a database table corresponding to a personalized query statement, the query semantics, and the query condition.

[0111] In one embodiment, see Figure 6 , the above-mentioned step S140 may include steps S141 to S144.

[0112] S141 . Classify and identify a database table of personalized data by using a similarity calculation method according to the personalized query sentence, so as to determine a database table corresponding to the personalized query sentence.

[0113] In this embodiment, the personalized database table is classified and judged, that is, it is determined which data table the current question needs to be queried from. Since there are multiple database tables for a scenario, such as the above-mentioned credit card scenario business, potential database tables are obtained according to the current business scenario, and similarity calculation is used to determine which database table description best matches the current question.

[0114] The features used in similar calculations are as follows: key words in the question, as well as the word parts and entity attributes of the words, for example, the word segmentation result of the question "How many times have I been overdue this year?" is "How many times have I been overdue this year?", as well as their word entity attributes, for example, for "overdue" in the semantic words there is "word category: table; attributed to one word: overdue information; corresponding database table: t_user overdue information table" and other information; the name of each database table, description information and database table field information, for example: for t_user overdue information table, its name: "t_user overdue information table", description information: overdue, number of overdue times, bad records, etc., field information includes: overdue time, overdue amount, etc.

[0115] The Cosine similarity calculation model based on deep vectorized features is used to convert the above questions and each library table into vectorized representations, and then deep feature extraction is performed through a multi-layer Bible network. Finally, the questions and the library table are converted into one-bit vectors of the same length through full connection. Finally, similarity calculation is performed by calculating the Cosine similarity between the two vectors, thereby determining that the above example question is querying the "t_User Overdue Information Table" database table.

[0116] S142: extract relevant fields from a database table corresponding to the personalized query sentence according to the personalized query sentence to obtain query semantics.

[0117] In this embodiment, the query semantics refers to the fields to be queried.

[0118] The semantic information of the sentences in the identified database table is obtained, for example, including fields such as user number, overdue time, overdue amount, etc. Since SQL query also has other SQL intrinsic support abstract processing data items such as querying the total number of records that meet the conditions count(*), querying the average value avg(x), maximum value max(x) of a data field, etc. Therefore, these SQL intrinsic supported processed field information are merged with the original field information, and the whole is used as a data candidate data item field set. For the database table "t_User Overdue Information Table" queried by the above question, the overall candidate fields involved become: [user number, overdue time, overdue amount, count(*)cnt, avg(overdue amount)avg_1, max(overdue amount)max_1, ...]

[0119] The goal is to identify the field items of the user's current actual query, so as to convert it into selecting one or more best semantically matching data field items from the above overall field list.

[0120] For the semantic matching of data field items, we use classification: we perform binary classification calculations based on the word segmentation recognition information of the user's question + the literal features and semantic description information of each data item, and if they meet the requirements, they are added to the query data field item. For the semantic features of data field items, for example: count(*) can be described as: count, number of times, total number of times, etc., and avg (overdue amount) can be described as: avg, average overdue amount, etc. In addition, in addition to some field information explicitly mentioned by the user, some outbound call scenarios require the configuration of default data information for certain information table queries. For example: for the credit card repayment information table, it is generally necessary to return the default amount of arrears, minimum repayment amount, repayment date and other field information.

[0121] In general, TextCNN and other methods are used to classify and calculate each field option to select and determine the current user query field data item, and the default field information is combined to confirm the overall query field. For the above example "How many times have I been overdue this year", the system can finally select [count(*)cnt] as the user query semantics.

[0122] S143: Identify the semantic incidental information in the personalized query sentence to obtain a query condition.

[0123] In this embodiment, the query condition refers to other limiting conditions during the query.

[0124] Automatic addition of default additional conditions such as user number. Since users will only query their own data when making outbound calls, it is necessary to query the current user's mobile phone number information to obtain the user's bank internal number information, and then add the user number information to the query conditions. Identification of semantic additional information in question queries. For example, for the above sentence "How many times have I been overdue this year?", we know through entity recognition that "this year" is a time entity. According to its entity link information, it is [2021-01-01 to 2021-12-31], and combined with the "overdue time" field in the database table, the value range of its field is also "time". Therefore, according to the rule matching method, the semantic information such as "overdue time" between "2021-01-01" and "2021-12-31" is obtained.

[0125] S144: Assemble the database table corresponding to the personalized query sentence, the query semantics, and the query condition to obtain a query SQL statement.

[0126] The database table to be identified and queried, the data field items to be retrieved, the query conditions and other information are assembled and spliced ​​into the user's personalized query SQL to form an SQL statement that can be directly queried and executed from the database table.

[0127] S150: Query corresponding content from the personalized data according to the query SQL statement to form a query result.

[0128] In this embodiment, the query result includes a no-data response result or a data response result. When the query result is a data response result, it also includes the corresponding data found in the query.

[0129] The executable query SQL semantics are obtained, and then the corresponding database table is queried. The query results are further converted into text through NLG technology, and finally the text is converted into telephone voice feedback to the user through TTS speech synthesis.

[0130] S160, determining whether the query result is a no-data response result;

[0131] S170: If the query result is not a no-data response result, encapsulate the data corresponding to the query result and output the encapsulated data.

[0132] Specifically, the data corresponding to the query result is converted into text through NLG generation technology, and the text is output through TTS speech synthesis technology.

[0133] S180: If the query result is a no data response result, output a voice message indicating no query result.

[0134] There are two situations in data query: data is found and data is not found. If data is not found, NLG generation is relatively simple. Generally, manually written speech text is randomly selected for feedback, and marketing guidance for outbound call scenarios is carried out by combining the upper and lower processes. If data is found, the following two methods are generally used to generate the corresponding text: rule template method and automatic summary method. For the rule template method, it is generally based on the database table name, field name, user gender and other information to be queried.<K,V> The candidate template is determined by the exact matching method, as shown in Table 6 below, where K is the key value composed of the joint splicing of the table name + table field + personalized information field, and V is the response text template. As shown in the first row of Table 6, K = User Overdue Information Table: Overdue Amount: Male, V = Mr. ${Surname}, I see that you are overdue ${overdue amount}...

[0135] Table 6 Candidate templates

[0136]

[0137]

[0138] For some queries, if there is no corresponding text response template, it is necessary to use the Encoder-Decoder end-to-end sequence generation technology. At the input end, the user's original question, question entity segmentation, database table name, data field name, query condition, final query result, user gender and other information are vectorized and feature modeled, and then a natural language text sentence is generated at the output end. The structure of the entire Encoder-Decoder model is as follows: Figure 6 shown.

[0139] The training corpus for the above modeling process is shown in Table 7.

[0140] Table 7. Model training corpus

[0141]

[0142] With the above corpus, we can train a user query response model. In the outbound call production environment, we can directly assemble the user input in a similar form to the above, and generate the response text through the model forward prediction.

[0143] In this embodiment, a system construction method for real-time query response to personalized data mentioned by users in a bank outbound telephone marketing scenario includes a method for preparing user personalized inquiry data in a marketing scenario, how to perform high-quality SQL query conversion based on user questions during a call, and how to effectively respond to the final query results and feedback to the user. By building a real-time data query system through this method, it is possible to effectively solve some personalized dynamic answer inquiry problems in actual marketing scenarios, improve communication fluency, and make it difficult for users to find that it is a "machine" that is influencing, thereby improving the overall marketing efficiency and conversion rate.

[0144] The above-mentioned outbound call scenario data query method builds personalized data from historical outbound call scenarios. When conducting a conversation, it performs question identification, question parsing and query on the conversation data, and encapsulates the query results and outputs them, thereby realizing the outbound call system's ability to perform real-time data query and providing accurate response results for personalized data query needs.

[0145] Figure 7 FIG. 3 is a schematic block diagram of an outbound call scene data query device 300 provided in an embodiment of the present invention. Figure 7 As shown, corresponding to the above outbound call scene data query method, the present invention also provides an outbound call scene data query device 300. The outbound call scene data query device 300 includes a unit for executing the above outbound call scene data query method, and the device can be configured in a server. Figure 7 The outbound call scenario data query device 300 includes a data construction unit 301, a conversation data acquisition unit 302, a processing unit 303, a parsing unit 304, a query unit 305, a judgment unit 306, an encapsulation unit 307 and an output unit 308.

[0146] The data construction unit 301 is used to construct real-time query data for various outbound call scenarios to obtain personalized data; the conversation data acquisition unit 302 is used to obtain the conversation data of the outbound call scenario; the processing unit 303 is used to perform question discrimination processing on the conversation data to obtain personalized query sentences; the parsing unit 304 is used to parse the personalized query sentences into query SQL statements; the query unit 305 is used to query the corresponding content from the personalized data according to the query SQL statement to form a query result; the judgment unit 306 is used to judge whether the query result is a no-data response result; the encapsulation unit 307 is used to encapsulate the data corresponding to the query result if the query result is not a no-data response result, and output the encapsulated data. The output unit 308 is used to output a voice of no query result if the query result is a no-data response result.

[0147] In one embodiment, the data construction unit 301 includes a history information recording subunit, a data item identification subunit, a merging subunit, and a table creation subunit.

[0148] The historical information recording subunit is used to record historical outbound call information according to business scenarios; the data item identification subunit is used to identify the personalized query data items of each conversation in the historical outbound call information according to set rules; the merging subunit is used to merge and count the personalized query data items of each conversation to obtain statistical results; the table creation subunit is used to construct relevant database tables for the statistical results according to the common real-time query data items of the set scenarios to obtain personalized data.

[0149] In one embodiment, the processing unit 303 includes an entity identification subunit, a link processing subunit, and a type identification subunit.

[0150] The entity recognition subunit is used to perform word segmentation and entity recognition on the dialogue data to obtain entity words; the link processing subunit is used to perform entity link processing on the entity words to obtain semantic information; the type recognition subunit is used to identify the question type of the semantic information to obtain a personalized query question.

[0151] In one embodiment, the link processing subunit is used to align the entity words with standardized semantic entities to obtain semantic information.

[0152] In one embodiment, the type identification subunit is used to identify the question type of the semantic information by using a rule and sequence labeling method or a TextCNN short text classification model to obtain a personalized query question.

[0153] In one embodiment, the parsing unit 304 includes a classification identification subunit, a field extraction subunit, an accompanying information identification subunit, and an assembly subunit.

[0154] A classification and identification subunit is used to classify and identify the database table of personalized data by similarity calculation according to the personalized query sentence, so as to determine the database table corresponding to the personalized query sentence; a field extraction subunit is used to extract relevant fields from the database table corresponding to the personalized query sentence according to the personalized query sentence, so as to obtain query semantics; an incidental information identification subunit is used to identify the semantic incidental information in the personalized query sentence, so as to obtain query conditions; an assembling subunit is used to assemble the database table corresponding to the personalized query sentence, the query semantics and the query conditions, so as to obtain a query SQL statement.

[0155] In one embodiment, the encapsulation unit 307 is used to convert the data corresponding to the query result into text using NLG technology, and output the text using TTS speech synthesis technology.

[0156] It should be noted that technicians in the relevant field can clearly understand that the specific implementation process of the above-mentioned outbound call scene data query device 300 and each unit can refer to the corresponding description in the aforementioned method embodiment, and for the convenience and brevity of description, it will not be repeated here.

[0157] The outbound call scene data query device 300 can be implemented in the form of a computer program. The computer program can be used in Figure 8 Runs on the computer device shown.

[0158] See also Figure 8 , Figure 8 5 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0159] See also Figure 8 The computer device 500 includes a processor 502 , a memory and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0160] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, and when the program instructions are executed, the processor 502 can execute an outbound call scenario data query method.

[0161] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500 .

[0162] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute an outbound call scenario data query method.

[0163] The network interface 505 is used to communicate with other devices over the network. Figure 8 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0164] The processor 502 is used to run the computer program 5032 stored in the memory to implement the following steps:

[0165] Construct real-time query data for various outbound call scenarios to obtain personalized data; obtain conversation data for outbound call scenarios; perform question discrimination processing on the conversation data to obtain personalized query sentences; parse the personalized query sentences into query SQL statements; query corresponding content from the personalized data according to the query SQL statements to form query results; determine whether the query result is a no-data response result; if the query result is not a no-data response result, encapsulate the data corresponding to the query result and output the encapsulated data.

[0166] In one embodiment, when the processor 502 implements the step of constructing real-time query data for various outbound call scenarios to obtain personalized data, the processor 502 specifically implements the following steps:

[0167] Record historical outbound call information according to business scenarios; identify personalized query data items for each conversation in the historical outbound call information according to set rules; merge and count the personalized query data items for each conversation to obtain statistical results; construct relevant database tables for the statistical results according to common real-time query data items in the set scenarios to obtain personalized data.

[0168] In one embodiment, when the processor 502 implements the step of performing question identification processing on the dialog data to obtain a personalized query sentence, the processor 502 specifically implements the following steps:

[0169] The conversation data is segmented and entity recognized to obtain entity words; the entity words are entity linked to obtain semantic information; the semantic information is question type recognized to obtain personalized query questions.

[0170] In one embodiment, when the processor 502 implements the step of performing entity linking processing on the entity words to obtain semantic information, the processor 502 specifically implements the following steps:

[0171] The entity words are aligned with standardized semantic entities to obtain semantic information.

[0172] In one embodiment, when the processor 502 implements the step of identifying the question type of the semantic information to obtain a personalized query sentence, the processor 502 specifically implements the following steps:

[0173] The semantic information is identified by using a rule and sequence labeling method or a TextCNN short text classification model to obtain a personalized query sentence.

[0174] In one embodiment, when the processor 502 implements the step of parsing the personalized query sentence into a query SQL statement, the processor 502 specifically implements the following steps:

[0175] According to the personalized query sentence, a database table of personalized data is classified and identified by a similarity calculation method to determine the database table corresponding to the personalized query sentence; according to the personalized query sentence, relevant fields are extracted from the database table corresponding to the personalized query sentence to obtain query semantics; semantic incidental information in the personalized query sentence is identified to obtain query conditions; the database table corresponding to the personalized query sentence, the query semantics and the query conditions are assembled to obtain a query SQL statement.

[0176] In one embodiment, when the processor 502 implements the step of encapsulating the data corresponding to the query result and outputting the encapsulated data, the processor 502 specifically implements the following steps:

[0177] The data corresponding to the query result is converted into text through NLG technology, and the text is output through TTS speech synthesis technology.

[0178] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit 303 (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0179] It can be understood by those skilled in the art that all or part of the processes in the method for implementing the above embodiment can be completed by instructing the relevant hardware through a computer program. The computer program includes program instructions, and the computer program can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiment of the above method.

[0180] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor executes the following steps:

[0181] Construct real-time query data for various outbound call scenarios to obtain personalized data; obtain conversation data for outbound call scenarios; perform question discrimination processing on the conversation data to obtain personalized query sentences; parse the personalized query sentences into query SQL statements; query corresponding content from the personalized data according to the query SQL statements to form query results; determine whether the query result is a no-data response result; if the query result is not a no-data response result, encapsulate the data corresponding to the query result and output the encapsulated data.

[0182] In one embodiment, when the processor executes the computer program to implement the step of constructing real-time query data for various outbound call scenarios to obtain personalized data, the processor specifically implements the following steps:

[0183] Record historical outbound call information according to business scenarios; identify personalized query data items for each conversation in the historical outbound call information according to set rules; merge and count the personalized query data items for each conversation to obtain statistical results; construct relevant database tables for the statistical results according to common real-time query data items in the set scenarios to obtain personalized data.

[0184] In one embodiment, when the processor executes the computer program to implement the step of performing question identification processing on the dialogue data to obtain a personalized query sentence, the processor specifically implements the following steps:

[0185] The conversation data is segmented and entity recognized to obtain entity words; the entity words are entity linked to obtain semantic information; the semantic information is question type recognized to obtain personalized query questions.

[0186] In one embodiment, when the processor executes the computer program to implement the step of performing entity linking processing on the entity words to obtain semantic information, the processor specifically implements the following steps:

[0187] The entity words are aligned with standardized semantic entities to obtain semantic information.

[0188] In one embodiment, when the processor executes the computer program to implement the step of identifying the question type of the semantic information to obtain a personalized query sentence, the processor specifically implements the following steps:

[0189] The semantic information is identified by using a rule and sequence labeling method or a TextCNN short text classification model to obtain a personalized query sentence.

[0190] In one embodiment, when the processor executes the computer program to implement the step of parsing the personalized query sentence into a query SQL statement, the processor specifically implements the following steps:

[0191] According to the personalized query sentence, a database table of personalized data is classified and identified by a similarity calculation method to determine the database table corresponding to the personalized query sentence; according to the personalized query sentence, relevant fields are extracted from the database table corresponding to the personalized query sentence to obtain query semantics; semantic incidental information in the personalized query sentence is identified to obtain query conditions; the database table corresponding to the personalized query sentence, the query semantics and the query conditions are assembled to obtain a query SQL statement.

[0192] In one embodiment, when the processor executes the computer program to implement the step of encapsulating the data corresponding to the query result and outputting the encapsulated data, the processor specifically implements the following steps:

[0193] The data corresponding to the query result is converted into text through NLG technology, and the text is output through TTS speech synthesis technology.

[0194] The storage medium may be a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk, etc., which are computer-readable storage media that can store program codes.

[0195] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0196] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and there may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0197] The steps in the method of the embodiment of the present invention can be adjusted in order, combined and deleted according to actual needs. The units in the device of the embodiment of the present invention can be combined, divided and deleted according to actual needs. In addition, the functional units in various embodiments of the present invention can be integrated into a processing unit 303, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0198] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, terminal, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0199] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. The outbound call scenario data query method is characterized by: include: Build real-time query data for various outbound call scenarios to obtain personalized data; Get the conversation data of the outbound call scenario; Performing question sentence discrimination processing on the dialogue data to obtain personalized query sentences; Parsing the personalized query into a query SQL statement; Querying corresponding content from the personalized data according to the query SQL statement to form a query result; Determine whether the query result is a no-data response result; If the query result is not a no-data response result, encapsulating the data corresponding to the query result and outputting the encapsulated data; The step of parsing the personalized query sentence into a query SQL statement includes: Classify and identify the database table of personalized data by using a similarity calculation method according to the personalized query sentence, so as to determine the database table corresponding to the personalized query sentence; Extracting relevant fields from a database table corresponding to the personalized query sentence according to the personalized query sentence to obtain query semantics; Identifying semantic incidental information in the personalized query sentence to obtain query conditions; The database table corresponding to the personalized query sentence, the query semantics and the query condition are assembled to obtain a query SQL statement.

2. The outbound call scenario data query method according to claim 1, characterized in that: The real-time query data for various outbound call scenarios is constructed to obtain personalized data, including: Record historical outbound call information according to business scenarios; Identify the personalized query data item of each conversation in the historical outbound call information according to the set rules; The personalized query data items of each conversation are combined and counted to obtain statistical results; The statistical results are used to construct relevant database tables according to common real-time query data items in the set scenarios to obtain personalized data.

3. The outbound call scenario data query method according to claim 1, characterized in that: The question identification processing of the dialogue data to obtain a personalized query sentence includes: Performing word segmentation and entity recognition on the conversation data to obtain entity words; Performing entity linking processing on the entity words to obtain semantic information; The question type is identified on the semantic information to obtain a personalized query sentence.

4. The outbound call scenario data query method according to claim 3, characterized in that: The performing entity linking processing on the entity words to obtain semantic information includes: The entity words are aligned with standardized semantic entities to obtain semantic information.

5. The outbound call scenario data query method according to claim 1, characterized in that: The step of identifying the question type of the semantic information to obtain a personalized query sentence includes: The semantic information is identified by using a rule and sequence labeling method or a TextCNN short text classification model to obtain a personalized query sentence.

6. The outbound call scenario data query method according to claim 1, characterized in that: The step of encapsulating the data corresponding to the query result and outputting the encapsulated data includes: The data corresponding to the query result is converted into text through NLG technology, and the text is output through TTS speech synthesis technology.

7. Outbound call scene data query device, characterized in that: include: Data construction unit, used to construct real-time query data for various outbound call scenarios to obtain personalized data; A conversation data acquisition unit, used to acquire conversation data of an outbound call scenario; A processing unit, configured to perform question identification processing on the dialogue data to obtain a personalized query sentence; A parsing unit, used to parse the personalized query sentence into a query SQL statement; A query unit, used to query corresponding content from the personalized data according to the query SQL statement to form a query result; A judging unit, used to judge whether the query result is a no-data response result; an encapsulation unit, configured to encapsulate data corresponding to the query result if the query result is not a no-data response result, and output the encapsulated data; The parsing unit includes a classification identification subunit, a field extraction subunit, an accompanying information identification subunit and an assembly subunit; A classification and identification subunit is used to classify and identify the database table of personalized data by similarity calculation according to the personalized query sentence, so as to determine the database table corresponding to the personalized query sentence; a field extraction subunit is used to extract relevant fields from the database table corresponding to the personalized query sentence according to the personalized query sentence, so as to obtain query semantics; The incidental information identification subunit is used to identify the semantic incidental information in the personalized query sentence to obtain the query condition; The assembling subunit is used to assemble the database table corresponding to the personalized query sentence, the query semantics and the query condition to obtain a query SQL statement.

8. A computer device, characterized in that: The computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 6 when executing the computer program.

9. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 6 can be implemented.

Images