Statement quality inspection method, device and system and storage medium

Abstract

The invention relates to the technical field of data analysis, and particularly provides a statement quality inspection method, device and system and a storage medium. According to the statement quality inspection method disclosed by the invention, the quality inspection rule is divided into the global quality inspection rule and the single-sentence quality inspection rule, the global quality inspection result is obtained based on the global quality inspection rule in each quality inspection period, and single-sentence quality inspection is performed in combination with cached quality inspection range information and historical quality inspection results to obtain single-sentence quality inspection results; the statement quality inspection result is obtained by integrating the global quality inspection result and the single statement quality inspection result, the global detection capability can be achieved, meanwhile, repeated quality inspection of historical statements is avoided, computing resource waste is reduced, the quality inspection efficiency is improved, and the robustness and the quality inspection effect of a quality inspection system are better.

Description

Technical Field

[0001] The present disclosure relates to the technical field of data analysis, and in particular to a sentence quality inspection method, device, system and storage medium. Background Art

[0002] At present, many industries rely on customer service to carry out business such as customer service, product sales, and after-sales service. The customer service's business capabilities directly affect the company's service quality. Therefore, quality inspection of customer service conversation sentences is an important part of improving the company's service quality and risk control capabilities.

[0003] In the related art, the quality inspection efficiency and accuracy of the real-time quality inspection solution for call scenarios are poor, and it is difficult to meet the requirements of the quality inspection scenario. Summary of the invention

[0004] In order to improve the efficiency and accuracy of sentence quality inspection, the embodiments of the present disclosure provide a sentence quality inspection method, device, system and storage medium.

[0005] In a first aspect, an embodiment of the present disclosure provides a sentence quality inspection method, comprising:

[0006] Determine the global quality inspection results of the current quality inspection cycle based on the global quality inspection rules;

[0007] Based on the single sentence quality inspection rules and the cached quality inspection range information, determine the inspection range of the current quality inspection cycle from the full set of sentences, where the full set of sentences includes the incremental sentences of all quality inspection cycles;

[0008] Performing single sentence quality inspection on the sentences within the detection range based on the single sentence quality inspection rule, and obtaining the single sentence quality inspection result of the current quality inspection cycle in combination with the cached historical quality inspection results;

[0009] The sentence quality inspection result of the current quality inspection cycle is determined according to the global quality inspection result and the single sentence quality inspection result.

[0010] In a second aspect, the present disclosure provides a sentence quality inspection device, including:

[0011] A global quality inspection module, configured to determine a global quality inspection result of a current quality inspection cycle based on a global quality inspection rule;

[0012] A scope determination module is configured to determine the detection scope of the current quality inspection cycle from the full amount of sentences based on the single sentence quality inspection rules and the cached quality inspection scope information, wherein the full amount of sentences includes the incremental sentences of all quality inspection cycles;

[0013] A single sentence quality inspection module is configured to perform single sentence quality inspection on the sentences within the detection range based on the single sentence quality inspection rule, and obtain the single sentence quality inspection result of the current quality inspection cycle in combination with the cached historical quality inspection results;

[0014] The result determination module is configured to determine the sentence quality inspection result of the current quality inspection cycle according to the global quality inspection result and the single sentence quality inspection result.

[0015] In some implementations, the single sentence quality inspection module is configured to:

[0016] Determine the quality inspection scope information of the current quality inspection cycle according to the detection scope of the current quality inspection cycle;

[0017] Based on the quality-inspected range information of the current quality-inspection cycle, the cached quality-inspected range information is updated, and based on the single-sentence quality-inspection result of the current quality-inspection cycle, the cached historical quality-inspection result is updated.

[0018] In some embodiments, the global quality inspection module is configured to:

[0019] In response to the time series identifier of the incremental statement of the current quality inspection cycle being earlier than the time series identifier of the incremental statement of the previous quality inspection cycle, the quality inspection operation of the current quality inspection cycle is abandoned.

[0020] In some embodiments, the result determination module is configured to:

[0021] In response to the statement quality inspection result meeting a preset condition, a prompt message is generated.

[0022] In a third aspect, the present disclosure provides a speech quality inspection system, including:

[0023] Processor; and

[0024] A memory storing computer instructions, wherein the computer instructions are used to enable the processor to execute the method described in any embodiment of the first aspect.

[0025] In a fourth aspect, an embodiment of the present disclosure provides a storage medium storing computer instructions, wherein the computer instructions are used to enable a computer to execute the method according to any embodiment of the first aspect.

[0026] The sentence quality inspection method of the embodiment of the present disclosure divides the quality inspection rules into global quality inspection rules and single-sentence quality inspection rules, and in each quality inspection cycle, obtains the global quality inspection result based on the global rules, performs single-sentence quality inspection in combination with the cached quality inspection range information and historical quality inspection results to obtain the single-sentence quality inspection result, and obtains the sentence quality inspection result by combining the global quality inspection result and the single-sentence quality inspection result. It can have global detection capabilities while avoiding repeated quality inspection of historical sentences, reducing the waste of computing resources, improving quality inspection efficiency, and improving the robustness of the quality inspection system and the quality inspection effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the specific embodiments of the present disclosure or the technical solutions in the prior art, the drawings required for use in the specific embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0028] Figure 1 It is an architectural diagram of a quality inspection system according to some embodiments of the present disclosure.

[0029] Figure 2 It is a flowchart of a sentence quality inspection method according to some implementation modes of the present disclosure.

[0030] Figure 3 It is a flowchart of a sentence quality inspection method according to some implementation modes of the present disclosure.

[0031] Figure 4 It is a flowchart of a sentence quality inspection method according to some implementation modes of the present disclosure.

[0032] Figure 5 It is a flowchart of a sentence quality inspection method according to some implementation modes of the present disclosure.

[0033] Figure 6 It is a flowchart of a sentence quality inspection method according to some implementation modes of the present disclosure.

[0034] Figure 7 It is a structural block diagram of a sentence quality inspection device in some implementation modes of the present disclosure.

[0035] Figure 8 It is a structural block diagram of a computer system for implementing the method disclosed in the present invention. DETAILED DESCRIPTION

[0036] The technical solution of the present disclosure will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described implementation is a part of the implementation of the present disclosure, rather than all the implementations. Based on the implementation in the present disclosure, all other implementations obtained by ordinary technicians in the field without creative work are within the scope of protection of the present disclosure. In addition, the technical features involved in the different implementations of the present disclosure described below can be combined with each other as long as they do not conflict with each other.

[0037] It should be noted that if the following disclosure involves user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.), all of them are information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0038] At present, many industries rely on telephones to conduct business such as customer service, product sales, and after-sales service. The business capabilities of manual customer service directly affect the company's service quality. Therefore, quality inspection of customer service conversation statements is an important part of improving the company's service quality and risk control capabilities.

[0039] The traditional method is to save the recordings of calls between customer service and customers, and then conduct quality inspection on the customer service statements through manual sampling. The manual quality inspection method is very inefficient, with low quality inspection coverage. The quality inspection standards will also change with the subjective tendencies of the quality inspectors, resulting in high costs and poor quality inspection results.

[0040] In recent years, with the development of NLP (Natural Language Processing) technology, more and more solutions have been proposed to use quality inspection models to perform sentence quality inspection on customer service voice, text and other content. For example, a detection model based on deep neural networks (DNN) can be used to replace manual work, and based on pre-established quality inspection rules, offline quality inspection of call recordings can be achieved, with higher quality inspection coverage and efficiency. However, this offline quality inspection solution can only be carried out after the conversation between the customer service and the customer is over, which has a lag and cannot pre-position risks. In other words, it is impossible to discover in real time during the conversation that the customer service has problems that affect user emotions, such as illegal language, irrelevant answers, negative communication, etc.

[0041] In the related technology, the real-time online quality inspection solutions for customer service dialogue scenarios mainly include incremental quality inspection and full quality inspection. Incremental quality inspection means that for every additional sentence or sentences in the dialogue scenario, the quality inspection rules are traversed for the newly added sentences, and the quality inspection results are fed back in real time. Full quality inspection means that for every additional sentence or sentences in the dialogue scenario, the quality inspection rules are traversed for all sentences or a whole paragraph of sentences in combination with the previous historical sentences, and the quality inspection results are fed back in real time.

[0042] For incremental quality inspection solutions, due to the lack of context detection capabilities, only a few basic quality inspection rules can be implemented, and quality inspection of complex rules and global rules cannot be implemented.

[0043] For example, in a real-time quality inspection scenario, a single sentence rule is "customer service volume cannot exceed X value". If the customer service volume exceeds X value, it means that the customer service may be emotional, which brings a bad communication experience to the user. However, in some scenarios, if the customer in the above text says "Can you speak louder? It's too noisy here and I can't hear clearly", then in actual scenarios, the customer service should be allowed to increase the volume. However, for the incremental quality inspection solution, it can only perform single sentence quality inspection on incremental sentences, and does not have the ability to combine context, so it cannot effectively respond to this scenario, resulting in the feedback quality inspection results misleading customer service, which in turn affects the user communication experience.

[0044] For example, in a customer service conversation scenario, after a user asks a question, they expect the customer service to answer as quickly as possible. Therefore, the customer service response time also directly affects the user's communication experience. However, the incremental quality inspection solution cannot process global rules (such as average response time), and thus cannot implement quality inspection for global rules.

[0045] As for the full-volume quality inspection solution, since it combines the full amount of sentences for quality inspection each time, it can effectively deal with the above-mentioned defects of incremental quality inspection. However, in the real-time quality inspection scenario, every time a sentence or a few sentences are added, all quality inspection rules must be traversed once for the current full amount of sentences. As the number of dialogue data increases, the system burden is heavy and the computational cost is extremely high. Moreover, each quality inspection will generate redundant detection of historical sentences, wasting computing resources and reducing quality inspection efficiency.

[0046] For example, in a real-time quality inspection scenario, a quality inspection rule may be "banned words exist in customer service statements". In a full-volume quality inspection solution, each time the quality inspection rule is called to perform quality inspection on all statements, the quality inspection rule will be traversed for all statements, and historical statements will be repeatedly quality inspected, wasting computing resources.

[0047] Based on the above, it can be seen that in the real-time sentence quality inspection solution of the related art, it is impossible to take into account both the quality inspection accuracy and the quality inspection efficiency, resulting in poor sentence quality inspection results. Therefore, the embodiments of the present disclosure provide a sentence quality inspection method, device, system and storage medium, which are intended to improve the quality inspection efficiency while taking into account the context detection capability, realize global rule detection, and effectively be compatible with missed calls, misaligned scenarios and offline detection capabilities.

[0048] Figure 1 The following is a diagram showing the architecture of the sentence quality inspection system in some embodiments of the present disclosure. Figure 1 The application scenarios of the disclosed solution are described.

[0049] like Figure 1 As shown, in the sentence quality inspection scenario disclosed in the present invention, a client 100, a customer service end 200 and a quality inspection system 300 are included.

[0050] The client 100 refers to an electronic device used by the customer, and the customer service end 200 refers to an electronic device used by the customer service. The client 100 and the customer service end 200 can establish a communicable connection through a wireless network to achieve communication in scenarios such as voice calls, video calls, and text chats. For example, taking a voice call as an example, the customer can send a speech voice (customer statement) to the customer service end 200 through the client, and the customer service receives the customer statement at the customer service end 200 and makes a corresponding reply. The customer service end 200 sends the customer service speech voice (customer service statement) to the client 100, so that the customer can receive the customer service statement at the client 100.

[0051] The quality inspection system 300 and the customer service end 200 can establish a communicable connection wirelessly or wiredly, so that the quality inspection system 300 can obtain real-time conversation data from the customer service end 200, and the conversation data includes customer statements and customer service statements, and then implement real-time quality inspection of the conversation data according to the statement quality inspection method described below in the present disclosure, and feed back the quality inspection results to the customer service end 200. The customer service end 200 can generate corresponding prompt information according to the feedback quality inspection results to remind the customer service to pay attention to relevant matters.

[0052] Understandably, Figure 1 The example only shows a conversation scenario between a group of clients 100 and a customer service end 200. In the embodiment of the present disclosure, the quality inspection system 300 can communicate with multiple customer service ends 200 at the same time. That is, the quality inspection system 300 can process conversation data between multiple groups of clients 100 and customer service ends 200 at the same time, and realize sentence quality inspection of multiple customer service agents.

[0053] Based on the understanding of the above sentence quality inspection scenario, the following Figure 2 The sentence quality inspection method of the embodiment of the present disclosure is described. In the embodiment of the present disclosure, the sentence quality inspection method can be Figure 1 The example quality inspection system performs processing.

[0054] like Figure 2 As shown, in some embodiments, the sentence quality inspection method of the present disclosure example includes:

[0055] S210: Determine the global quality inspection result of the current quality inspection cycle based on the global quality inspection rule.

[0056] It can be understood that for the real-time statement quality inspection scenario, as the conversation between the customer service and the customer proceeds, statements are continuously generated in an incremental manner, so the real-time statement quality inspection process will be carried out periodically.

[0057] However, it is worth noting that the quality inspection cycle is not limited in the embodiments of the present disclosure. For example, in some embodiments, a quality inspection may be performed every predetermined time period, so that the quality inspection cycle is divided by time; for example, in other embodiments, a quality inspection may be performed every time one or more sentences are added, so that the quality inspection cycle is divided by the number of sentences. Those skilled in the art can understand this, and the present disclosure will not elaborate on it.

[0058] In the disclosed implementation manner, sentence quality inspection rules are divided into global quality inspection rules and single sentence quality inspection rules in advance according to different quality inspection rules.

[0059] It is understandable that in the sentence quality inspection scenario, since the content of the conversation is very complex, the quality inspection rules that need to be formulated are also very complex. Some quality inspection rules are rules for global sentence quality inspection, such as global average speaking speed, global average reply speed, etc., while some quality inspection rules are rules for single sentence quality inspection, such as keyword detection, volume detection, etc.

[0060] In the traditional incremental real-time quality inspection solution, it can only implement quality inspection of basic single-sentence rules, and cannot cover global quality inspection rules. In the traditional full-volume real-time quality inspection solution, its quality inspection rules do not distinguish between global rules and single-sentence rules. Each quality inspection is to traverse all quality inspection rules for all sentences, resulting in redundant detection and wasting computing resources. For ease of understanding, the present disclosure provides a simple example below.

[0061] For example, in a real-time sentence quality inspection scenario, the quality inspection rules are as follows:

[0062] Rule 1: Keyword k1 is identified in sentence x, where sentence x refers to the xth sentence.

[0063] Rule 2: In statement x, it is recognized that the volume exceeds the value t.

[0064] Rule 3: Keyword k1 is recognized in statement x, and keyword k2 is recognized in statement [x, x+3]. [x, x+3] refers to statements x to x+3, i.e., statement x and the three statements after it.

[0065] Rule 4: The average speaking speed of customer service personnel exceeds the value s;

[0066] Rule 5: The average time customers wait for customer service replies exceeds the T value.

[0067] Rule 6:

[0068] In this example scenario, if the traditional incremental real-time quality inspection solution is used, it can only process quality inspection rules for single sentences, such as "Rule 1" and "Rule 2", because it can only perform quality inspection on newly added sentences. Quality inspection cannot be performed on "Rule 3" due to the lack of contextual integration capability. Quality inspection cannot be performed on "Rule 4" and "Rule 5" due to the lack of quality inspection capability for all sentences.

[0069] If the traditional full-scale real-time quality inspection solution is adopted, although rules 1 to 5 can be covered, each quality inspection requires traversing rules 1 to 5 for all statements. For historical statements, they need to be repeatedly inspected by quality inspection rules 1 to 3 during each quality inspection process, resulting in a huge waste of resources and reducing quality inspection efficiency.

[0070] Therefore, in the embodiments of the present disclosure, the quality inspection rules can be divided in advance, and the quality inspection rules can be divided into global quality inspection rules and single sentence quality inspection rules according to the scope of the sentences covered by the quality inspection rules. Global quality inspection rules refer to rules for quality inspection of all sentences, and single sentence quality inspection rules refer to rules for quality inspection of single sentences.

[0071] For example, in the above example, rules 1 to 3 are all rules that need to be quality checked for a single sentence, so rules 1 to 3 are the single sentence quality check rules described in this disclosure. Rules 4 and 5 are rules that need to be quality checked for all sentences, so rules 4 to 5 are the global quality check rules described in this disclosure.

[0072] It is worth noting that each quality inspection rule includes one or more quality inspection points, and each quality inspection point refers to a basic quality inspection sub-rule.

[0073] A relatively simple quality inspection rule generally includes only one basic quality inspection sub-rule, that is, only one quality inspection point. For example, Rule 1, Rule 2, Rule 4 and Rule 5 in the above example only include one quality inspection point.

[0074] For more complex quality inspection rules, they will include multiple basic quality inspection sub-rules, that is, multiple quality inspection points, and multiple quality inspection points can have dependencies. For example, Rule 3 in the above example includes two basic quality inspection sub-rules: "keyword k1 is recognized in sentence x" and "keyword k2 is recognized in sentence [x, x+3]". That is, Rule 3 includes two quality inspection points, namely:

[0075] Quality control point P1: keyword k1 is recognized in sentence x;

[0076] Quality inspection point P2: The keyword k2 is recognized in the sentence [x, x+3].

[0077] Moreover, the quality inspection point P2 depends on the quality inspection point P1, that is, the statement detection range of the quality inspection point P2 [x, x+3] depends on the statement x that hits the quality inspection point P1. Of course, those skilled in the art can understand that a certain quality inspection rule may also include more quality inspection points, and the dependency relationship between the quality inspection points may also be more complicated, which will not be elaborated in this disclosure.

[0078] Therefore, in the embodiment of the present disclosure, the quality inspection rules are divided into global quality inspection rules and single sentence quality inspection rules according to the sentence scope covered by the quality inspection rules in advance, and the global quality inspection rules include one or more global quality inspection points, and the single sentence quality inspection rules include one or more single sentence quality inspection points.

[0079] In the real-time quality inspection scenario, taking a quality inspection cycle as an example, the quality inspection cycle will generate one or more new statements, which are the incremental statements described in this disclosure. At the same time, historical statements may also be included before the quality inspection cycle. For example, in one example, statements S3, S4, and S5 are added in the current quality inspection cycle, then the newly added statements S3 to S5 are the incremental statements of the current quality inspection cycle, and the previously generated S1 to S2 are historical statements.

[0080] In S210 of the disclosed embodiment, quality inspection is performed on all sentences based on pre-divided global quality inspection rules to obtain global quality inspection results.

[0081] The full statement includes the incremental statements of the current quality inspection cycle and the historical statements generated before. For example, in the above example, the current quality inspection cycle adds statements S3, S4 and S5, and the historical statements S1 to S2 are combined with the incremental statements S3 to S5 to obtain the full statements S1 to S5. Of course, it can be understood that if no historical statements were generated before the current quality inspection cycle, the full statement corresponding to the current quality inspection cycle only includes the incremental statements of the current quality inspection cycle.

[0082] After determining the full statement corresponding to the current quality inspection cycle, the full statement can be quality inspected using the global quality inspection rules to obtain the corresponding global quality inspection results. For example, in some implementations, the full statement can be traversed through all global quality inspection points included in the global quality inspection rules, and then the corresponding global quality inspection results can be obtained. The following implementations of this disclosure illustrate this.

[0083] S220: Based on the single sentence quality inspection rules and the cached quality inspection range information, determine the inspection range of the current quality inspection cycle from the entire sentence.

[0084] It can be understood that for global quality inspection rules, the range of sentences that need to be quality inspected is all sentences. For single sentence quality inspection rules, in order to avoid repeated quality inspection of historical sentences, in the embodiment of the present disclosure, a cache mechanism is used to cache the quality inspection range information of sentences that have been quality inspected in each quality inspection cycle, so that in the next quality inspection cycle, the inspection range that needs to be inspected for a single sentence in the current quality inspection cycle can be determined based on the quality inspection range information cached in the previous quality inspection cycle.

[0085] As can be seen from the above, the quality inspection rules may include multiple single-sentence quality inspection rules, and each single-sentence quality inspection rule includes one or more single-sentence quality inspection points. For example, in the above example, all quality inspection rules include three single-sentence quality inspection rules, namely Rule 1, Rule 2, and Rule 3, where Rule 1 and Rule 2 each include one single-sentence quality inspection point, and Rule 3 includes two single-sentence quality inspection points.

[0086] In the embodiments of the present disclosure, the quality-inspected range information cached in each quality inspection cycle includes the quality-inspected range information corresponding to each single-sentence quality inspection point, that is, the quality-inspected range information corresponding to each single-sentence quality inspection point is stored separately in the quality inspection system. In some embodiments, as the dialogue data is generated, corresponding identification information can be assigned to each sentence in chronological order. The identification information can be, for example, S1, S2, S3, S4..., and the quality-inspected range information of each single-sentence quality inspection point stored in the quality inspection system includes the identification information of these sentences.

[0087] Therefore, in the current quality inspection cycle, for each single sentence quality inspection point, the detection range corresponding to the single sentence quality inspection point in the current quality inspection cycle can be determined according to the quality inspection sub-rule of the single sentence quality inspection point and the quality inspection range information cached in the previous quality inspection cycle.

[0088] It can be understood that the quality-checked range information cached in the previous quality-check cycle refers to the range of sentences that have been quality-checked before the current quality-check cycle. In order to avoid repeated quality-checking of these quality-checked sentences in the current quality-check cycle, these quality-checked sentences can be removed from the sentences that need to be quality-checked in the current quality-check cycle, and only incremental single-sentence quality-checking is performed on the un-quality-checked sentences.

[0089] For ease of understanding, taking the single sentence quality inspection rule of the above rule 1 as an example, in the last quality inspection cycle, sentences S1 and S2 have been quality inspected, so the quality inspection range information cached in the last quality inspection cycle includes {S1, S2}. In the current quality inspection cycle, new sentences S3, S4 and S5 are added, and the current full sentence includes {S1, S2, S3, S4, S5}. In the current quality inspection cycle, the quality inspection sentences can be removed from the full sentence {S1, S2, S3, S4, S5} according to the cached quality inspection sentence range {S1, S2} to obtain the detection range {S3, S4, S5} of the current quality inspection cycle.

[0090] It is worth noting that in the disclosed embodiment, using single sentence quality inspection rules and cache mechanism to determine the detection scope of each quality inspection cycle is not equivalent to directly performing single sentence quality inspection on incremental sentences. This is because the rules corresponding to single sentence quality inspection points are different, and the scope of sentences that need to be inspected is also different. Not all single sentence quality inspection points are for incremental sentences.

[0091] For example, taking the "Rule 3" mentioned in the previous article as an example, Rule 3 includes single sentence quality check point P1 and single sentence quality check point P2. Single sentence quality check point P2 does not need to perform quality check on all single sentences, but only needs to perform quality check on sentence x that hits single sentence quality check point P1 and the three sentences after x.

[0092] In the example scenario described above, statements S1 and S2 have been quality checked for single sentences in the last quality inspection cycle. Assuming that neither of them hits the single sentence quality inspection point P1, the quality checked statement range corresponding to the single sentence quality inspection point P2 is empty. In the current quality inspection cycle, statements S3, S4, and S5 are added, and the current full set of statements includes {S1, S2, S3, S4, S5}. In the current quality inspection cycle, assuming that statement S4 hits the single sentence quality inspection point P1, according to the rules of the single sentence quality inspection point P2, the statements that need to be quality checked for the single sentence quality inspection point P2 include S4 and S5. At the same time, combined with the quality inspection range information (empty set) corresponding to the single sentence quality inspection point P2 cached in the last quality inspection cycle, it is determined that the detection range corresponding to the single sentence quality inspection point P2 in the current quality inspection cycle is {S4, S5}.

[0093] From this example, we can see that although the incremental sentences of the current quality inspection cycle include {S3, S4, S5}, the sentence detection range for the single sentence quality inspection point P2 only includes {S4, S5}. Therefore, if the single sentence quality inspection rules based on the division are used to perform single sentence quality inspection on the incremental sentences of each quality inspection cycle, it can only be applied to relatively basic rules, and cannot effectively deal with complex rules including multiple single sentence quality inspection points, because not all single sentence quality inspection points are for incremental sentences.

[0094] In the embodiments of the present disclosure, by utilizing the quality-inspected range information cached in each quality inspection cycle, the detection range corresponding to each single-sentence quality inspection point can be accurately calculated based on the rules of each single-sentence quality inspection point and the quality-inspected range information. This can not only meet the quality inspection scenarios with complex rules, but also avoid unnecessary quality inspections, fully utilize computing resources, and improve quality inspection efficiency.

[0095] The detection scope of the current quality inspection cycle obtained by S220 may include the detection scope corresponding to each single sentence quality inspection point included in the single sentence quality inspection rule. The detection scope of a single sentence quality inspection point means that the sentences within the detection scope need to be quality inspected according to the quality inspection sub-rule of the single sentence quality inspection point. For example, in the previous example, the single sentence quality inspection rule includes single sentence quality inspection point P1 and single sentence quality inspection point P2, then the detection scope of the current quality inspection cycle determined includes: the detection scope of single sentence quality inspection point P1, and the detection scope of single sentence quality inspection point P2.

[0096] S230 , performing single sentence quality inspection on the sentences within the detection range based on the single sentence quality inspection rules, and obtaining the single sentence quality inspection result of the current quality inspection cycle in combination with the cached historical quality inspection results.

[0097] In the disclosed embodiment, in each quality inspection cycle, not only the quality inspection scope information of each single sentence quality inspection point in the quality inspection cycle needs to be cached, but also the quality inspection results of each single sentence quality inspection point need to be cached, that is, the historical quality inspection results include the quality inspection results corresponding to each single sentence quality inspection point.

[0098] For example, in one example, the single-sentence quality inspection rule only includes one single-sentence quality inspection point. In the last quality inspection cycle, single-sentence quality inspection was performed on sentences S1 and S2, and the quality inspection result was {S2}. The quality inspection result indicates that sentence S2 hits the single-sentence quality inspection point, so the quality inspection range information cached in the last quality inspection cycle includes {S1, S2}, and the cached historical quality inspection results include {S2}.

[0099] In the current quality inspection cycle, after the detection scope of the single sentence quality inspection rule is determined through S220, the sentences within the detection scope can be quality inspected according to the single sentence quality inspection rule, and the single sentence quality inspection result of the current quality inspection cycle is obtained by combining the historical quality inspection results cached in the previous quality inspection cycle.

[0100] Still using the above example for explanation, in the current quality inspection cycle, newly added statements S3, S4 and S5 are added. Through the above process, it is determined that the detection range of the single sentence quality inspection point in the current quality inspection cycle includes {S3, S4, S5}. Then, quality inspection can be performed on the detection range including statements S3, S4 and S5 based on the single sentence quality inspection point. Assuming that statements S3, S4 and S5 are all missed, the result of this quality inspection is an empty set, and the historical quality inspection result cached in the previous quality inspection cycle is {S2}. Therefore, the historical quality inspection result is taken as the union of the current quality inspection result, and the single sentence quality inspection result in the current quality inspection cycle is {S2}.

[0101] It can be seen that in the embodiments of the present disclosure, the use of a cache mechanism to cache the historical quality inspection results of each quality inspection cycle can fully combine the context detection capability and effectively deal with certain repetitive quality inspection rules. For example, the rule corresponding to a single sentence quality inspection point is "the number of times the keyword k3 is recognized is greater than or equal to 2 times." In the embodiments of the present disclosure, for each quality inspection cycle, the historical quality inspection results of the previous quality inspection cycle can be combined, so that the incremental quality inspection solution can be used to achieve the quality inspection capability of the context and achieve coverage of repetitive rules.

[0102] Of course, it can be understood that the above examples are all illustrated by taking a single-sentence quality inspection rule including only one single-sentence quality inspection point as an example. For complex single-sentence quality inspection rules including multiple single-sentence quality inspection points, the principle is the same as above. This disclosure will illustrate this with specific examples below and will not be described in detail here.

[0103] In the implementation of the present disclosure, after obtaining the detection range and single sentence quality inspection results of the current quality inspection cycle, the quality inspection range information and historical quality inspection results cached in the previous quality inspection cycle can be further updated for the next quality inspection cycle. This is explained in the following implementation of the present disclosure and will not be described in detail here.

[0104] S240: Determine the sentence quality inspection result of the current quality inspection cycle according to the global quality inspection result and the single sentence quality inspection result.

[0105] In the disclosed embodiment, after obtaining the global quality inspection result of the current quality inspection cycle through S210 and the single sentence quality inspection result of the current quality inspection cycle through S230, the global quality inspection result and the single sentence quality inspection result can be combined to determine the final sentence quality inspection result of the current quality inspection cycle.

[0106] It can be understood that the global quality inspection result is the result obtained by performing global rule quality inspection on all sentences, and the single sentence quality inspection result is the result obtained by performing single sentence rule quality inspection on sentences within the required range. The sentence quality inspection results obtained by combining the two can not only reflect the quality inspection effect of all sentences, but also have the efficiency of incremental quality inspection.

[0107] Compared with the traditional incremental real-time quality inspection solution, the disclosed implementation mode can realize the detection capability of all sentences because it is compatible with global quality inspection rules. Moreover, in each quality inspection cycle, by combining the quality inspection scope information of the previous quality inspection cycle and the historical quality inspection results, it can not only effectively filter out the quality inspection sentences, avoid repeated quality inspection and waste of computing resources, but also have the detection capability of context, covering more and more complex quality inspection rules.

[0108] Compared with the traditional full-scale real-time quality inspection solution, in each quality inspection cycle, the quality inspection scope information and historical quality inspection results of the previous quality inspection cycle can be combined, thereby screening out a large number of redundant quality inspection operations, reducing computing overhead and improving quality inspection efficiency.

[0109] From the above, it can be seen that in the embodiment of the present disclosure, the quality inspection rules are divided into global quality inspection rules and single-sentence quality inspection rules, and in each quality inspection cycle, the quality inspection scope information of the previous quality inspection cycle and the historical quality inspection results are combined to perform single-sentence quality inspection. This can not only have global detection capabilities, but also avoid repeated quality inspection of historical sentences, reduce the waste of computing resources, and improve quality inspection efficiency. In addition, it can also achieve coverage of more complex quality inspection rules, effectively deal with scenarios such as missed calls and misalignment, and the robustness of the quality inspection system and the quality inspection effect are stronger.

[0110] In some embodiments, in combination Figure 1 As shown, the conversation data sent by the client end 200 and acquired by the quality inspection system is often audio data in voice format, so the quality inspection system can convert the audio data into text data to obtain conversation data in text format.

[0111] In addition, in the process of converting the audio data in voice format into the dialogue data in text format, the word segmentation and sentence segmentation algorithms in the related technology can be used to divide the dialogue data in text format into single sentences, which are also the sentences described in this specification.

[0112] There are many algorithms related to speech-to-text conversion and word or sentence segmentation of text data. Those skilled in the art can undoubtedly understand and fully implement them by referring to the relevant technologies, and the present disclosure will not elaborate on them.

[0113] like Figure 3 As shown, in some implementations, the sentence quality inspection method of the present disclosure example, the process of determining the global quality inspection result includes:

[0114] S310: For the jth global quality inspection sub-rule, based on the jth global quality inspection sub-rule, perform global quality inspection on all sentences to obtain a first global inspection result of the jth global quality inspection sub-rule in the current quality inspection cycle.

[0115] S320: Determine a global quality inspection result according to the first global inspection result of the M global quality inspection sub-rules.

[0116] In the disclosed embodiment, the full statement includes the incremental statement added in the current quality inspection cycle and the historical statement generated previously. For example, in one example, the incremental statement added in the current quality inspection cycle includes {S3, S4, S5}, and the historical statement generated previously includes {S1, S2}, then the combined full statement includes {S1, S2, S3, S4, S5}.

[0117] According to the foregoing, the global quality inspection rule may include at least one global quality inspection point, and each global quality inspection point corresponds to a global quality inspection sub-rule. In the embodiment of the present disclosure, the global quality inspection rule may include a total of M global quality inspection points, and each global quality inspection point represents a global quality inspection sub-rule for quality inspection based on the entire statement, where M is an integer greater than or equal to 1.

[0118] For example, in some implementations, the rule corresponding to the global quality inspection point may be any one or more of the following rules, such as the global average speech rate, the global average reply speed, the global average volume, and the global average emotion value.

[0119] The global average speaking speed refers to the ratio of the number of words in all sentences to the time, which is used to reflect the speed of customer service dialogue. The global average reply speed refers to the average time between the end of the customer's sentence and the beginning of the customer service reply, which is used to reflect the speed of customer service answering. The global average volume refers to the average volume corresponding to all sentences, which is used to reflect whether the customer service volume is stable. The global average emotion value refers to the emotion value corresponding to the customer service sentence, which is used to reflect whether the customer service emotion is stable.

[0120] After the full statement is determined, the global quality check points can be traversed for the full statement to obtain a first global detection result corresponding to each global quality check point, where the first global detection result indicates whether the global quality check point is hit in the current full statement.

[0121] In the disclosed embodiment, for the jth global quality inspection point, the quality inspection sub-rule corresponding to the jth global quality inspection point can be used to quality inspect all sentences to obtain the first global inspection result of the jth global quality inspection point in the current quality inspection cycle, where 1≤j≤M.

[0122] For example, in one example, the global quality inspection rule includes M=2 global quality inspection points, and the full statement can be traversed through these two global quality inspection points respectively, and the first global detection result corresponding to each global quality inspection point can be obtained. For example, if the full statement hits a global quality inspection point, the first global detection result corresponding to the global quality inspection point is expressed as true (hit), otherwise it is expressed as false (missed).

[0123] Regarding the algorithm process of performing quality inspection on all sentences based on global quality inspection points, those skilled in the art can undoubtedly implement it according to specific global quality inspection rules and refer to relevant technologies. This disclosure cannot list them all, and only an example is used below to illustrate.

[0124] For example, in one example, the global quality inspection rule includes 1 global quality inspection point, and the rule corresponding to the global quality inspection point is "average speaking speed is greater than threshold 1". In each quality inspection cycle, the total duration and total number of words corresponding to all sentences can be counted, and the average speaking speed can be calculated using the ratio of total number of words to total duration. The calculated average speaking speed is then compared with threshold 1. If the average speaking speed is greater than threshold 1, it means that the global quality inspection result is hit, otherwise, it means that the global quality inspection result is not hit.

[0125] After obtaining the first global detection result corresponding to each global quality inspection point, the first global detection results of all M global quality inspection points can be integrated to obtain the global quality inspection result for all sentences.

[0126] For example, in one example, the global quality inspection rule is: if any global quality inspection point hits, then the global quality inspection result hits. Therefore, the global quality inspection result can be expressed as: t1 = (Q1 or Q2 or ... or Qn), where Qn represents the first global inspection result corresponding to the nth global quality inspection point, and t1 represents the global quality inspection result obtained for the quality inspection of all sentences in the current quality inspection cycle.

[0127] The above describes the process of determining the global quality inspection result t1 of the current quality inspection cycle, and the following describes the process of determining the single sentence quality inspection result t2 of the current quality inspection cycle. For ease of understanding, the following will describe the sentence quality inspection method of the disclosed example in conjunction with a specific real-time sentence quality inspection scenario.

[0128] In an example scenario, the global quality inspection rule includes one global quality inspection point Q1, and the global quality inspection sub-rule corresponding to the global quality inspection point Q1 is: the global average speech speed is greater than the s value. This global quality inspection sub-rule can be understood as the speaking speed of the customer service conversation cannot be too slow. If it is too slow, it will consume the customer's patience and affect the customer's mood. Therefore, the global average speech speed can be selected to be greater than a certain value.

[0129] In some embodiments, the single sentence quality inspection rule includes two single sentence quality inspection points, namely a first single sentence quality inspection point and a second single sentence quality inspection point. The single sentence quality inspection sub-rule corresponding to the first single sentence quality inspection point is: the single sentence statement hits the first preset rule; the single sentence quality inspection sub-rule corresponding to the second single sentence quality inspection point is: the context statement of the statement that hits the first preset rule hits the second preset rule.

[0130] For example, in an example, the first single sentence quality check point and the second single sentence quality check point are:

[0131] The first single sentence quality check point P1: keyword k1 is recognized in sentence x;

[0132] The second single sentence quality inspection point P2: the keyword K2 is recognized in the sentence [x, x+5].

[0133] This rule can be understood as, when a sentence is identified to contain keyword k1, keyword k2 is identified in the next 5 sentences. There are many scenarios corresponding to this rule. For example, in one example scenario, the customer says "What is the name of your company?" The customer expects to get the "company name" as soon as possible, so the customer service is required to say "XXX Co., Ltd." within 5 sentences. For example, in another example scenario, the customer says "What is the annual fee for your company's product A?" The customer expects to get the "annual fee price" as soon as possible, so the customer service is required to say "The annual fee for our product A is XX yuan / year" within 5 sentences. For example, in another example scenario, it is stipulated that if the customer service says the illegal words again within 5 sentences after saying the illegal words for the first time, it will be considered a violation.

[0134] Of course, those skilled in the art can understand that in the embodiments of the present disclosure, the global quality inspection rules and the single sentence quality inspection rules are not limited to the above examples, and the number of quality inspection points they each include can be greater. For example, in one example, the single sentence quality inspection rule can also include a third single sentence quality inspection point P3: the keyword K3 is recognized in the sentence [x-5, x], which can be understood as the keyword K3 is recognized in the previous 5 sentences when a sentence contains the keyword k1. The embodiments of the present disclosure cannot be exhaustive in this regard. For the sake of convenience in the following description, only the above simplified example scenario is used as an example to illustrate the sentence quality inspection method of the embodiments of the present disclosure.

[0135] 1) Initial quality inspection cycle T 0

[0136] Initial quality inspection cycle T 0 Indicates the first quality check cycle of real-time statement quality check, at which time the system cache is empty.

[0137] In the initial quality inspection cycle T 0 Add statements S1 and S2. At this time, since there is no historical statement, the full statement is {S1, S2}.

[0138] In this inspection cycle T 0 In Figure 3 The method process of the implementation method performs quality inspection on the entire sentence {S1, S2} based on the global quality inspection point Q1 to obtain the global quality inspection result t1 corresponding to the global quality inspection point Q1, and this process will not be described in detail. In one example, it is assumed that the global quality inspection result t1 = true.

[0139] For the single sentence quality inspection rule, due to the initial quality inspection cycle T 0 The system cache is empty, therefore, for the first single sentence quality check point P1, its detection range is {S1, S2}, and the rules of the first single sentence quality check point P1 are used to perform quality checks on sentences S1 and S2, that is, to detect whether keyword k1 is recognized in sentences S1 and S2.

[0140] In an example, assuming that sentence S2 hits the first single sentence quality inspection point P1, that is, keyword k1 is recognized in sentence S2, then the quality inspection range information A1 corresponding to the first single sentence quality inspection point P1 is {S1, S2}, and the quality inspection result a1 is {S2}.

[0141] In the embodiment of the present disclosure, the quality inspection range information A1={S1, S2} and the quality inspection result a1={S2} corresponding to the first single sentence quality inspection point P1 can be cached in the system memory as the initial quality inspection cycle T 0 The single sentence quality inspection result corresponding to the first single sentence quality inspection point P1 in .

[0142] For the second single sentence quality inspection point P2, according to its quality inspection rules, the second single sentence quality inspection point P2 needs to detect the sentence x that hits the first single sentence quality inspection point P1 and the 5 sentences after sentence x. In the above example, the sentence that hits the first single sentence quality inspection point P1 is S2, and the current latest sentence is also S2. Therefore, the detection range A2 = {S2} corresponding to the second single sentence quality inspection point P2 can be determined.

[0143] Then, the sentence S2 is quality checked using the quality check rule of the second single sentence quality check point P2, that is, whether the keyword k2 is recognized in the sentence S2. In one example, assuming that the sentence S2 does not hit the second single sentence quality check point P2, that is, the keyword k2 is not recognized in the sentence S2, then the quality check range information A2 corresponding to the second single sentence quality check point P2 = {S2}, and the quality check result a2 = {}.

[0144] In the embodiment of the present disclosure, the quality inspection range information A2={S2} corresponding to the second single sentence quality inspection point P2 and the quality inspection result a2={} are cached in the system memory as the initial quality inspection cycle T 0 The single sentence quality inspection result corresponding to the second single sentence quality inspection point P2 in .

[0145] Through the above example process, in the initial quality inspection cycle T 0 After the completion, the system cache data can be shown in the following table 1:

[0146] Table 1. Quality inspection cycle T 0 Caching Data

[0147] First sentence quality check point P1 Second sentence quality inspection point P2 Information on the scope of quality inspection S1, S2 S2 Historical quality inspection results S2 None

[0148] Thus, in this example scenario, the first single sentence quality inspection point P1 is hit, and the second single sentence quality inspection point P2 is not hit, so the single sentence quality inspection result t2 is expressed as: t2 = (P1 and P2) = (true and false) = false, indicating that the initial quality inspection cycle T 0 The quality inspection result of the single sentence t2 was not hit.

[0149] Through the above example process, we can get the initial quality inspection cycle T 0 The global quality inspection result t1 = true, and the single sentence quality inspection result t2 = false, so the initial quality inspection cycle T 0 The corresponding sentence quality inspection result is expressed as: t = (t1 and t2) = (true and false) = false, which means that in the initial quality inspection period T 0 The corresponding statement quality inspection result is not hit, and the initial quality inspection cycle T is completed 0 quality inspection process.

[0150] 2) Quality inspection cycle T 1

[0151] In the embodiment of the present disclosure, the quality inspection cycle T 1 is the initial quality inspection cycle T 0 The next quality inspection cycle, assuming that in the quality inspection cycle T 1 In the new statement S3, S4 and S5, the following will be based on the quality inspection cycle T 1 As the current quality inspection cycle, the quality inspection method of the disclosed sentences is explained.

[0152] First, through Figure 3 The implementation process can determine the quality inspection cycle T 1 That is, according to the incremental statements {S3, S4, S5} and the historical statements {S1, S2}, the full statements {S1, S2, S3, S4, S5} can be determined, and then the full statements {S1, S2, S3, S4, S5} are quality checked based on the global quality check point Q1 to obtain the global quality check result t1' corresponding to the global quality check point Q1. Those skilled in the art can refer to Figure 3 The method process can undoubtedly be implemented and will not be described in detail here.

[0153] In one example, assuming that the quality inspection cycle T 1 The global quality inspection result t1'=true.

[0154] Then, it is necessary to combine the previous quality inspection cycle (quality inspection cycle T 0 ) cached quality inspection range information to determine the quality inspection point of each single sentence in the current quality inspection cycle (quality inspection cycle T 1 ) in the corresponding detection range, the following is combined with Figure 4 The implementation method is described.

[0155] like Figure 4 As shown, in some embodiments, the sentence quality inspection method of the present disclosure example, the process of determining the inspection scope of the current quality inspection cycle includes:

[0156] S410: For the i-th single sentence quality inspection sub-rule, determine the to-be-detected range information of the i-th single sentence quality inspection sub-rule from the entire sentence.

[0157] S420: Determine the detection scope of the i-th single sentence quality inspection sub-rule in the current quality inspection cycle according to the to-be-detected scope information and the quality inspection completed scope information of the i-th single sentence quality inspection sub-rule.

[0158] For the single sentence quality inspection point P1, according to the quality inspection sub-rules of the single sentence quality inspection point P1, the single sentence quality inspection point P1 needs to perform quality inspection on each sentence to determine whether the keyword k1 is recognized in the sentence. Therefore, the range information A3 to be detected corresponding to the single sentence quality inspection point P1 is the full sentence {S1, S2, S3, S4, S5}, which is expressed as the range information to be detected A3 = {S1, S2, S3, S4, S5}.

[0159] However, as mentioned above, in the last quality inspection cycle T 0 In the example, sentences S1 and S2 have been quality checked according to the single sentence quality check point P1. To avoid repeated quality checks, the quality checked sentences need to be filtered out and only the sentences that have not been quality checked are retained.

[0160] In the disclosed embodiment, the quality-checked range information A1={S1, S2} corresponding to the first single sentence quality-check point P1 can be read from the cache shown in Table 1, which indicates that the sentence has been quality-checked at the first single sentence quality-check point P1.

[0161] Therefore, according to the to-be-detected range information A3 = {S1, S2, S3, S4, S5} and the inspected range information A1 = {S1, S2}, the current inspection period T can be obtained. 1 The detection range A4 that needs to be detected by the first single sentence quality inspection point P1 is expressed as A4=A3-A1={S3, S4, S5}.

[0162] Similarly, for the single sentence quality inspection point P2, according to the quality inspection sub-rule of the single sentence quality inspection point P2, the single sentence quality inspection point P2 needs to perform quality inspection on the sentence that hits the single sentence quality inspection point P1 and the 5 sentences after the sentence, so as to determine whether the keyword k2 is recognized in the sentence. Combined with the above Table 1, it can be seen that the sentence S2 hits the first single sentence quality inspection point P1, so the to-be-detected range information A5 corresponding to the single sentence quality inspection point P2 = {S2, S3, S4, S5}.

[0163] However, as mentioned above, in the last quality inspection cycle T 0 In the example, sentence S2 has been quality checked according to single sentence quality check point P2. To avoid repeated quality checks, it is necessary to filter out the quality checked sentences and only keep the sentences that have not been quality checked.

[0164] In the disclosed embodiment, the inspected range information A2={S2} corresponding to the second single sentence inspection point P2 can be read from the cache shown in Table 1, which indicates that the sentence has been inspected at the second single sentence inspection point P2.

[0165] Thus, according to the to-be-detected range information A5={S2, S3, S4, S5} and the inspected range information A2={S2}, the current inspection period T can be obtained. 1 The detection range A6 that needs to be detected by the second single sentence quality inspection point P2 is expressed as A6=A5-A2={S3, S4, S5}.

[0166] Through the above process, the detection range A4 = {S3, S4, S5} of the first single sentence quality inspection point P1 and the detection range A6 = {S3, S4, S5} of the second single sentence quality inspection point P2 are obtained respectively. It can be seen that in the embodiment of the present disclosure, based on the quality inspection range information cached in the previous quality inspection cycle, repeated quality inspections can be effectively avoided, computing resources can be effectively utilized, and quality inspection efficiency can be improved.

[0167] After determining the detection range of each single sentence quality inspection point, the sentences within the sentence detection range can be quality inspected based on the quality inspection rules of the single sentence quality inspection point, thereby obtaining the single sentence quality inspection results of the current quality inspection cycle. Figure 5 Provide explanation.

[0168] like Figure 5 As shown, in some implementations, the sentence quality inspection method of the present disclosure example, the process of obtaining the single sentence quality inspection result of the current quality inspection cycle includes:

[0169] S510: For the i-th single sentence quality inspection sub-rule, perform single sentence quality inspection on the sentences within the inspection range to obtain the first single sentence detection result of the i-th single sentence quality inspection sub-rule in the current quality inspection cycle.

[0170] S520. Obtain a target detection result corresponding to the i-th single sentence quality inspection sub-rule according to the first single sentence detection result and the historical single sentence detection results.

[0171] S530 , obtaining a single sentence quality inspection result according to the target detection results of the N single sentence quality inspection sub-rules.

[0172] In the disclosed implementation manner, the historical single sentence detection result is the quality inspection result corresponding to each single sentence quality inspection point cached in the previous quality inspection cycle.

[0173] In the above example, for the first single sentence quality inspection point P1, through the above Figure 4 The method process has obtained its corresponding detection range A4 = {S3, S4, S5}, so that the sentences S3, S4, and S5 can be quality checked according to the quality inspection sub-rule of the first single sentence quality inspection point P1, that is, whether the keyword k1 is recognized in the sentences S3, S4, and S5, and the current quality inspection period T is obtained. 1 The first single sentence detection result a3 for the quality inspection of the detection range A4. For example, in one example, the first single sentence detection result a3 = {}, that is, the keyword k1 is not recognized in the sentences S3, S4 and S5.

[0174] At the same time, as shown in Table 1, in the last quality inspection cycle T 0 , the historical quality inspection result a1 corresponding to the first single sentence quality inspection point P1 = {S2}, and the historical quality inspection result a1 is the historical single sentence detection result corresponding to the first single sentence quality inspection point P1 described in the present disclosure.

[0175] Therefore, for the first single sentence quality inspection point P1, according to the current quality inspection cycle T 1 The first single sentence detection result a3 = {} in the previous quality inspection cycle T 0 The historical single sentence detection result a1={S2} in the above formula can obtain the target detection result a4=a3∪a1={S2} corresponding to the first single sentence quality inspection point P1, indicating that sentence S2 hits the first single sentence quality inspection point P1, thus in the current quality inspection cycle T 1 The first single sentence quality inspection point P1 is hit.

[0176] Similarly, for the second single sentence quality inspection point P2, through the above Figure 4 The method process has obtained its corresponding detection range A6 = {S3, S4, S5}, so that the sentences S3, S4, and S5 can be quality checked according to the quality inspection sub-rule of the second single sentence quality inspection point P2, that is, whether the keyword k2 is recognized in the sentences S3, S4, and S5, and the current quality inspection period T is obtained. 1 The first single sentence detection result a5 for the sentence detection range A6 is checked. For example, in one example, the first single sentence detection result a5 = {S4}, that is, the keyword k2 is recognized in the sentence S4.

[0177] At the same time, as shown in Table 1, in the last quality inspection cycle T 0 , the historical quality inspection result a2 corresponding to the second single sentence quality inspection point P1 = {}, and the historical quality inspection result a2 is the historical single sentence detection result corresponding to the second single sentence quality inspection point P2 described in the present disclosure.

[0178] Therefore, for the second single sentence quality inspection point P2, according to the current quality inspection cycle T 1The first single sentence detection result a5 = {S4} in the previous quality inspection cycle T 0 The second single sentence detection result a2 = {} in the result, the target detection result a6 = a5 ∪ a2 = {S4} corresponding to the second single sentence quality inspection point P2 can be obtained, indicating that the sentence S4 hits the second single sentence quality inspection point P2, thus in the current quality inspection cycle T 1 The second sentence quality inspection point P2 is hit.

[0179] Through the above process, we obtain the target detection result a4={S2} of the first single sentence quality inspection point P1 and the target detection result a6={S4} of the second single sentence quality inspection point P2, which means that in the current quality inspection cycle T 1 In the example, the first single sentence quality check point P1 hits (true), and the second single sentence quality check point P2 hits (true). Thus, according to the target detection results of the two single sentence quality check points, the current quality check period T can be obtained. 1 The single sentence quality inspection result t2' in is expressed as t2' = (P1 and P2) = (true and true) = true, that is, the current quality inspection cycle T 1 The quality inspection result of the single sentence is hit.

[0180] In summary, through the aforementioned Figure 3 The example process has obtained the current quality inspection cycle T 1 The global quality inspection result t1'=true in Figure 4 , Figure 5 The example process has obtained the current quality inspection cycle T 1 The single sentence quality inspection result t2'=true in the current quality inspection cycle T 1 The corresponding sentence quality inspection result t is expressed as: t = (t1'and t2') = (true and true) = true, indicating that the current quality inspection cycle T 1 The corresponding statement quality inspection result is a hit.

[0181] It is worth noting that in the embodiments of the present disclosure, the specific algorithm process of sentence quality inspection based on the single sentence quality inspection point can undoubtedly be implemented by those skilled in the art according to specific rules and with reference to relevant technologies. The present disclosure cannot be exhaustive in this regard, and only one example is used below to illustrate. For example, taking the first single sentence quality inspection point P1 as an example, when performing quality inspection on a certain sentence, a keyword detection algorithm can be used to identify the sentence, thereby determining whether the sentence contains preset keywords.

[0182] Through the above process, the current quality inspection cycle T can be completed. 1 The statement quality inspection process, in the quality inspection cycle T 1 After the statement quality inspection process is completed, it is also necessary to 1The quality inspection results of the previous quality inspection cycle T 0 The cached data is updated, combined with Figure 6 Provide explanation.

[0183] like Figure 6 As shown, in some implementations, the sentence quality inspection method of the example of the present disclosure further includes:

[0184] S610: Determine the quality inspection scope information corresponding to the current quality inspection cycle according to the detection scope of the current quality inspection cycle.

[0185] S620: Based on the quality-inspected range information corresponding to the current quality-inspection cycle, the cached quality-inspected range information is updated, and based on the single-sentence quality-inspection result corresponding to the current quality-inspection cycle, the cached historical quality-inspection result is updated.

[0186] For the first single sentence quality inspection point P1, Figure 4 and Figure 5 The method process has obtained its quality inspection period T 1 The corresponding detection range A4 = {S3, S4, S5}, and the target detection result a4 = {S2}. Thus, combined with the detection range A4 = {S3, S4, S5}, and the last quality inspection cycle T 0 The quality inspection range information A1={S1, S2} corresponding to the first single sentence quality inspection point P1 in the cache can be obtained to obtain the current quality inspection period T 1 The corresponding quality inspection range information A7 = A4 ∪ A1 = {S1, S2, S3, S4, S5}, the quality inspection range information A7 can be understood as: in the quality inspection cycle T 1 After completion, the sentence range of the first single sentence quality inspection point P1 has been quality inspected.

[0187] Similarly, for the second single sentence quality inspection point P2, through the above Figure 4 and Figure 5 The method process has obtained its quality inspection period T 1 The corresponding detection range A6 = {S3, S4, S5}, and the target detection result a6 = {S4}. Thus, combined with the detection range A6 = {S3, S4, S5}, and the last quality inspection cycle T 0 The quality inspection range information A2={S2} corresponding to the cached second single sentence quality inspection point P2 can be obtained to obtain the current quality inspection period T 1 The corresponding quality inspection range information A8 = A6 ∪ A2 = {S2, S3, S4, S5}, the quality inspection range information A8 can be understood as: in the quality inspection cycle T 1 After completion, the sentence range of the second single sentence quality inspection point P2 has been quality inspected.

[0188] Thus, according to the quality-inspected range information A7 corresponding to the first single-sentence quality inspection point P1 and the quality-inspected range information A8 corresponding to the second single-sentence quality inspection point P2, the quality-inspected range information shown in Table 1 is updated. In addition, according to the target detection result a4 corresponding to the first single-sentence quality inspection point P1 and the target detection result a6 corresponding to the second single-sentence quality inspection point P2, the historical quality inspection results shown in Table 1 are updated. In this example, the updated cache data can be shown in Table 2 below:

[0189] Table 2. Quality inspection cycle T 1 Cached data

[0190]

[0191] The data cached in Table 2 can be shown in the quality inspection cycle T 1 In the example, the sentences that have been quality checked at the first single sentence quality check point P1 include S1 to S5, and the corresponding hit item is sentence S2; the sentences that have been quality checked at the second single sentence quality check point P2 include S2 to S5, and the corresponding hit item is S4.

[0192] 3) Quality inspection cycle T 2

[0193] In the embodiment of the present disclosure, the quality inspection cycle T 2 is the quality inspection cycle T 1 The next quality inspection cycle, assuming that in the quality inspection cycle T 2 New statements S6 and S7 are added in the quality inspection cycle T 2 The quality inspection process is completely based on the quality inspection cycle T 1 Those skilled in the art can undoubtedly understand and fully implement the above embodiments with reference to the above embodiments, and the present disclosure will not elaborate on this.

[0194] Through the above process, the sentence quality inspection of each quality inspection cycle can be realized, and the cached data of each quality inspection cycle can be updated so that the next quality inspection cycle can call the cached data of the previous quality inspection cycle, realize the incremental quality inspection of single sentences, avoid repeated quality inspection of past sentences, and improve the quality inspection efficiency. Moreover, during the quality inspection process of each quality inspection cycle, the global quality inspection rules can be combined to conduct quality inspection on all sentences, so as to have context detection capabilities and cover complex quality inspection rules.

[0195] Through the detection of each quality inspection cycle, until the conversation ends and no new sentences are generated, the entire customer service conversation real-time quality inspection process ends. In some embodiments of the present disclosure, all data cached during the quality inspection process can be cleared at the end of the quality inspection process.

[0196] It is worth noting that the sentence quality inspection method of the embodiment of the present disclosure, in addition to having the above-mentioned beneficial effects, can also be well compatible with missed calls, misalignment and offline quality inspection scenarios.

[0197] Combination Figure 1 As shown, in the real-time quality inspection scenario, as incremental statements are continuously generated, in each quality inspection cycle, the client 200 will assemble statements based on the incremental statements of the current quality inspection cycle and historical statements to obtain full statements, and then the client 200 will push the full statements and quality inspection requests of the current quality inspection cycle to the quality inspection system 300, so that the quality inspection system can perform statement quality inspection on the quality inspection statements based on global quality inspection rules and / or single sentence quality inspection rules. Thus, in each quality inspection cycle, the quality inspection system 300 implements statement quality inspection through the method process in any of the aforementioned implementation methods in response to the full statements and quality inspection requests pushed by the client 200, and returns the statement quality inspection results.

[0198] However, in some scenarios, affected by factors such as network fluctuations, the quality inspection requests received by the quality inspection system 300 may be omitted or misplaced, which will be explained below.

[0199] 1. Missed Call Scenario

[0200] The missed call scenario refers to a phenomenon in which, due to factors such as network fluctuations, requests are lost when the client 200 pushes sentence quality inspection requests to the quality inspection system 300 in real time.

[0201] For example, in one example, in the quality inspection cycle T 0 Add sentences S1 and S2, so that the client 200 pushes all sentences S1 and S2 and the quality inspection cycle T to the quality inspection system 300 0 The quality inspection system 300 completes the quality inspection cycle T based on the above method process. 0 The statement quality is checked and the statement quality check result is returned.

[0202] Continue real-time dialogue, during the quality inspection cycle T 1 New statements S3 and S4 are added, so that the client 200 assembles the full set of statements S1 to S4 based on the new statements S3 and S4 and the historical statements S1 and S2, and pushes the full set of statements S1 to S4 and the quality inspection cycle T to the quality inspection system 300. 1 However, due to factors such as network fluctuations, the quality inspection system 300 did not receive the quality inspection cycle T 1 The quality inspection request, so the quality inspection cycle T 1 If there is any missed inspection, the statement quality inspection result will not be returned.

[0203] Continue real-time dialogue, during the quality inspection cycle T 2 Add sentences S5 and S6, so that the client 200 assembles the full set of sentences S1 to S6 based on the new sentences S5 and S6 and the historical sentences S1 to S4, and pushes the full set of sentences S1 to S6 and the quality inspection cycle T to the quality inspection system 300.2 It is understandable that in the quality inspection cycle T 2 Since the data cached in the last quality inspection cycle is still the quality inspection cycle T 0 Therefore, based on the above method, the statements S3 and S4 that have not been quality checked can also be quality checked, and the correct statement quality check results can be returned. At the same time, based on the quality check cycle T 2 The statement quality inspection results will update the cached data to the latest status.

[0204] It can be seen that in the embodiments of the present disclosure, even if missed calls occur in a quality inspection cycle, it will not have any impact on the real-time quality inspection process, and accurate sentence quality inspection results can still be returned, and the quality inspection system has good robustness.

[0205] 2. Dislocation scene

[0206] The misalignment scenario refers to the situation where, due to factors such as network fluctuations and data asynchrony, the client 200 does not push statement quality inspection requests to the quality inspection system 300 in real time, but the requests do not arrive at the quality inspection system in the correct order, or the statement quality inspection results of the quality inspection system 300 are not returned to the client in the correct order.

[0207] For example, in one example, in the quality inspection cycle T 0 Add sentences S1 and S2, so that the client 200 pushes all sentences S1 and S2 and the quality inspection cycle T to the quality inspection system 300 0 The quality inspection system 300 completes the quality inspection cycle T based on the above method process. 0 The statement quality is checked and the statement quality check result is returned.

[0208] Continue real-time dialogue, during the quality inspection cycle T 1 New statements S3 and S4 are added, so that the client 200 assembles the full set of statements S1 to S4 based on the new statements S3 and S4 and the historical statements S1 and S2, and pushes the full set of statements S1 to S4 and the quality inspection cycle T to the quality inspection system 300. 1 Quality inspection request.

[0209] Continue real-time dialogue, during the quality inspection cycle T 2 Add sentences S5 and S6, so that the client 200 assembles the full set of sentences S1 to S6 based on the new sentences S5 and S6 and the historical sentences S1 to S4, and pushes the full set of sentences S1 to S6 and the quality inspection cycle T to the quality inspection system 300. 2 Quality inspection request.

[0210] However, due to factors such as network fluctuations and data asynchrony, the quality inspection cycle T 2 The quality inspection request is before the quality inspection period T 1 Arrival quality inspection system.

[0211] In some embodiments, the quality inspection system can compare the time stamp of the quality inspection request of the current quality inspection cycle with the time stamp of the quality inspection request of the previous quality inspection cycle. If the time stamp of the quality inspection request of the current quality inspection cycle is earlier than the time stamp of the quality inspection request of the previous quality inspection cycle, it means that the quality inspection request of the current quality inspection cycle is misaligned with the quality inspection request of the previous quality inspection cycle, and the quality inspection operation of the current quality inspection cycle can be directly abandoned.

[0212] For example, in the above example, due to the quality inspection cycle T 2 The quality inspection request is before the quality inspection period T 1 Arrives at the quality inspection system, so the quality inspection system has already targeted the quality inspection cycle T 2 The quality inspection request has completed the quality inspection, and the quality inspection cycle T has been fully covered 1 Therefore, for the quality inspection cycle T 1 The quality inspection request can be directly discarded to save computing resources.

[0213] For example, in another example, the quality inspection system 300 completes the quality inspection cycle T based on the aforementioned quality inspection method. 0 The quality inspection operation is performed, and the statement quality inspection result C1 is returned to the client 200. The real-time quality inspection is continued, and the quality inspection system 300 completes the quality inspection cycle T based on the aforementioned quality inspection method. 1 The quality inspection operation is performed, and the statement quality inspection result C2 is returned to the client 200. The real-time quality inspection is continued, and the quality inspection system 300 completes the quality inspection cycle T based on the aforementioned quality inspection method. 2 The quality inspection operation is performed and the statement quality inspection result C3 is returned to the client 200. However, due to factors such as network fluctuations and data asynchrony, the quality inspection cycle T 2 The statement quality inspection result C3 is earlier than the quality inspection cycle T 1 The statement quality inspection result C2 is returned to the client.

[0214] From the above, we can see that due to the quality inspection cycle T 2 The statement quality inspection result C3 can completely cover the quality inspection cycle T 1 The sentence quality inspection result is C2, so the client directly discards the quality inspection cycle T according to the time stamp. 1 The statement quality inspection result C2 is sufficient, and only the quality inspection period T is retained 2 The sentence quality inspection result is C3.

[0215] It can be seen that in the embodiments of the present disclosure, even if the quality inspection request or quality inspection result is misplaced, it will not have any impact on the real-time quality inspection process, and accurate statement quality inspection results can still be obtained, and the quality inspection system has good robustness.

[0216] 3) Offline scenario

[0217] The offline scenario refers to the scenario where the complete conversation recording between the customer service representative and the customer is saved and then quality checked.

[0218] The statement quality inspection method of the embodiment of the present disclosure is also compatible with offline quality inspection scenarios. Its method process is exactly the same as the aforementioned online quality inspection process, which is equivalent to quality inspecting all statements in one quality inspection cycle and then returning the quality inspection results.

[0219] It can be seen from the above that in the embodiments of the present disclosure, the quality inspection system can effectively be compatible with various scenarios such as real-time online quality inspection, offline quality inspection, missed calls, and misalignment, and is more robust.

[0220] In some implementations, in a real-time quality inspection scenario, after determining the sentence quality inspection result of the current quality inspection cycle, the sentence quality inspection method of the example of the present disclosure further includes:

[0221] In response to the statement quality inspection result meeting the preset condition, an alarm prompt message is generated.

[0222] Combination Figure 1 As shown, in the real-time quality inspection scenario, the statements of each quality inspection cycle are quality inspected by the statement quality inspection method described above. After obtaining the statement quality inspection result, if the statement quality inspection result meets the preset conditions, it means that the statement quality inspection result is hit, then the corresponding alarm prompt information can be generated in real time on the customer service end 200 to remind the customer service that there may be violations, please correct them in time to reduce risks.

[0223] For example, in the above example implementation, the initial quality inspection period T 0 If the sentence quality inspection result t=false is obtained, that is, it does not meet the preset conditions, then there is no need to generate an alarm prompt message. However, in the quality inspection cycle T1, if the sentence quality inspection result t=true is obtained, that is, it meets the preset conditions, then an alarm prompt message needs to be generated.

[0224] In the embodiments of the present disclosure, the alarm prompt information may be any type of information with a prompt function. For example, in one example, the alarm prompt information may be text information displayed on the display interface of the customer service terminal 200, such as "The current wording is illegal!" "Timeout for replying to the customer!" "The current mood fluctuates too much!" etc. Of course, those skilled in the art can understand that the alarm prompt information may also be other types of information, such as sound and light alarms, etc., and the present disclosure does not limit this.

[0225] From the above, it can be seen that in the embodiment of the present disclosure, the quality inspection rules are divided into global quality inspection rules and single-sentence quality inspection rules, and in each quality inspection cycle, the single-sentence quality inspection is performed in combination with the quality inspection sentence range of the previous quality inspection cycle and the historical quality inspection results. This can not only have global detection capabilities, but also avoid repeated quality inspection of historical sentences, reduce the waste of computing resources, and improve quality inspection efficiency. In addition, it can also achieve coverage of more complex quality inspection rules, effectively deal with scenarios such as missed calls, misalignment, and offline quality inspection, and the quality inspection system has stronger robustness and quality inspection effects.

[0226] In some embodiments, the present disclosure provides a sentence quality inspection device, which can be applied to, for example, Figure 1 The quality inspection system 300 is shown.

[0227] like Figure 7 As shown, in some embodiments, the sentence quality inspection device of the present disclosure example includes:

[0228] A global quality inspection module 10, configured to determine a global quality inspection result of a current quality inspection cycle based on global quality inspection rules;

[0229] A scope determination module 20 is configured to determine the detection scope of the current quality inspection cycle from the full amount of sentences based on the single sentence quality inspection rules and the cached quality inspection scope information, wherein the full amount of sentences includes the incremental sentences of all quality inspection cycles;

[0230] The single sentence quality inspection module 30 is configured to perform single sentence quality inspection on the sentences within the detection range based on the single sentence quality inspection rules, and obtain the single sentence quality inspection results of the current quality inspection cycle in combination with the cached historical quality inspection results;

[0231] The result determination module 40 is configured to determine the sentence quality inspection result of the current quality inspection cycle according to the global quality inspection result and the single sentence quality inspection result.

[0232] In some implementations, the single-sentence quality inspection rule includes N single-sentence quality inspection sub-rules, where N is an integer greater than or equal to 1; the cached quality-inspected range information includes the quality-inspected range information of the i-th single-sentence quality inspection sub-rule, where 1≤i≤N; and the range determination module 20 is configured as follows:

[0233] For the i-th single sentence quality inspection sub-rule, determine the to-be-detected range information of the i-th single sentence quality inspection sub-rule from the entire sentence;

[0234] The detection scope of the i-th single sentence quality inspection sub-rule in the current quality inspection cycle is determined according to the to-be-detected range information and the quality inspection range information of the i-th single sentence quality inspection sub-rule.

[0235] In some implementations, the historical quality inspection result includes the historical single sentence detection result of the i-th single sentence quality inspection sub-rule; the single sentence quality inspection module 30 is configured as follows:

[0236] For the i-th single sentence quality inspection sub-rule, perform single sentence quality inspection on the sentences within the detection range to obtain the first single sentence detection result of the i-th single sentence quality inspection sub-rule in the current quality inspection cycle;

[0237] Obtaining a target detection result corresponding to the i-th single sentence quality inspection sub-rule according to the first single sentence detection result and the historical single sentence detection results;

[0238] The single sentence quality inspection result is obtained according to the target detection results of N single sentence quality inspection sub-rules.

[0239] In some implementations, the single-sentence quality inspection rule includes at least two single-sentence quality inspection sub-rules, wherein the to-be-detected range information of at least one single-sentence quality inspection sub-rule depends on the target detection result of another single-sentence quality inspection sub-rule.

[0240] In some implementations, the global quality inspection rule includes M global quality inspection sub-rules, where M is an integer greater than or equal to 1; the global quality inspection module 10 is configured as follows:

[0241] The determining of the global quality inspection result of the current quality inspection cycle based on the global quality inspection rule includes:

[0242] For the jth global quality inspection sub-rule, based on the jth global quality inspection sub-rule, perform a global quality inspection on the entire sentence to obtain a first global inspection result of the jth global quality inspection sub-rule in the current quality inspection cycle, 1≤j≤M;

[0243] The global quality inspection result is determined according to the first global inspection results of the M global quality inspection sub-rules.

[0244] In some implementations, the global quality inspection sub-rule includes any one or more rules based on the global average speech rate, the global average reply speed, the global average volume, and the global average emotion value.

[0245] In some implementations, the single sentence quality inspection module 30 is configured to:

[0246] Determine the quality inspection scope information of the current quality inspection cycle according to the detection scope of the current quality inspection cycle;

[0247] Based on the quality-inspected range information of the current quality-inspection cycle, the cached quality-inspected range information is updated, and based on the single-sentence quality-inspection result of the current quality-inspection cycle, the cached historical quality-inspection result is updated.

[0248] In some embodiments, the global quality inspection module 10 is configured as follows:

[0249] In response to the time series identifier of the incremental statement of the current quality inspection cycle being earlier than the time series identifier of the incremental statement of the previous quality inspection cycle, the quality inspection operation of the current quality inspection cycle is abandoned.

[0250] In some implementations, the result determination module 40 is configured to:

[0251] In response to the statement quality inspection result meeting a preset condition, a prompt message is generated.

[0252] From the above, it can be seen that in the embodiment of the present disclosure, the quality inspection rules are divided into global quality inspection rules and single-sentence quality inspection rules, and in each quality inspection cycle, the single-sentence quality inspection is performed in combination with the quality inspection sentence range of the previous quality inspection cycle and the historical quality inspection results. This can not only have global detection capabilities, but also avoid repeated quality inspection of historical sentences, reduce the waste of computing resources, and improve quality inspection efficiency. In addition, it can also achieve coverage of more complex quality inspection rules, effectively deal with scenarios such as missed calls, misalignment, and offline quality inspection, and the quality inspection system has stronger robustness and quality inspection effects.

[0253] In some embodiments, the present disclosure provides a sentence quality inspection system, which can refer to, for example, the aforementioned Figure 1 In the quality inspection system 300 shown, in some examples, the sentence quality inspection system includes:

[0254] Processor; and

[0255] The memory stores computer instructions, where the computer instructions are used to enable the processor to execute the method of any of the aforementioned embodiments.

[0256] In some embodiments, the present disclosure provides a storage medium storing computer instructions, where the computer instructions are used to cause a computer to execute the method according to any of the foregoing embodiments.

[0257] Figure 8 A schematic diagram of the structure of a computer system 600 suitable for implementing the method disclosed in the present invention is shown. Figure 8 The system shown can realize the corresponding functions of the above-mentioned processor and quality inspection system 300.

[0258] like Figure 8 As shown, the computer system 600 includes a processor 601, which can perform various appropriate actions and processes according to the program stored in the memory 602 or the program loaded from the storage part 608 into the memory 602. Various programs and data required for the operation of the system 600 are also stored in the memory 602. The processor 601 and the memory 602 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0259] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed, so that a computer program read therefrom is installed into the storage section 608 as needed.

[0260] In particular, according to an embodiment of the present disclosure, the above method process can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program tangibly contained on a machine-readable medium, and the computer program includes a program code for executing the above method. In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 609, and / or installed from a removable medium 611.

[0261] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and a module, a program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0262] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the embodiments. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A sentence quality inspection method, characterized in that: include: Determine the global quality inspection results of the current quality inspection cycle based on the global quality inspection rules; Based on the single sentence quality inspection rules and the cached quality inspection range information, determine the inspection range of the current quality inspection cycle from the full set of sentences, where the full set of sentences includes the incremental sentences of all quality inspection cycles; Performing single sentence quality inspection on the sentences within the detection range based on the single sentence quality inspection rule, and obtaining the single sentence quality inspection result of the current quality inspection cycle in combination with the cached historical quality inspection results; The sentence quality inspection result of the current quality inspection cycle is determined according to the global quality inspection result and the single sentence quality inspection result.

2. The method according to claim 1, characterized in that The single sentence quality inspection rule includes N single sentence quality inspection sub-rules, where N is an integer greater than or equal to 1; the cached quality inspection range information includes the quality inspection range information of the i-th single sentence quality inspection sub-rule, where 1≤i≤N; The method of determining the inspection scope of the current inspection cycle from all sentences based on the single sentence inspection rules and the cached inspection scope information includes: For the i-th single sentence quality inspection sub-rule, determine the to-be-detected range information of the i-th single sentence quality inspection sub-rule from the entire sentence; The detection scope of the i-th single sentence quality inspection sub-rule in the current quality inspection cycle is determined according to the to-be-detected range information and the quality inspected range information of the i-th single sentence quality inspection sub-rule.

3. The method according to claim 2, characterized in that The historical quality inspection result includes the historical single sentence detection result of the i-th single sentence quality inspection sub-rule; the single sentence quality inspection is performed on the sentences within the detection range based on the single sentence quality inspection rule, and the single sentence quality inspection result of the current quality inspection cycle is obtained in combination with the cached historical quality inspection results, including: For the i-th single sentence quality inspection sub-rule, perform single sentence quality inspection on the sentences within the detection range to obtain the first single sentence detection result of the i-th single sentence quality inspection sub-rule in the current quality inspection cycle; Obtaining a target detection result of the i-th single sentence quality inspection sub-rule according to the first single sentence detection result and the historical single sentence detection results; The single sentence quality inspection result is obtained according to the target detection results of the N single sentence quality inspection sub-rules.

4. The method according to claim 3, characterized in that The single-sentence quality inspection rule includes at least two single-sentence quality inspection sub-rules, wherein the to-be-detected range information of at least one single-sentence quality inspection sub-rule depends on the target detection result of another single-sentence quality inspection sub-rule.

5. The method according to claim 1, characterized in that The global quality inspection rule includes M global quality inspection sub-rules, where M is an integer greater than or equal to 1; The determining of the global quality inspection result of the current quality inspection cycle based on the global quality inspection rule includes: For the jth global quality inspection sub-rule, based on the jth global quality inspection sub-rule, perform a global quality inspection on the entire sentence to obtain a first global inspection result of the jth global quality inspection sub-rule in the current quality inspection cycle, 1≤j≤M; The global quality inspection result is determined according to the first global inspection results of the M global quality inspection sub-rules.

6. The method according to claim 5, characterized in that The global quality inspection sub-rules include any one or more rules based on the global average speech speed, the global average reply speed, the global average volume, and the global average emotion value.

7. The method according to claim 1, characterized in that Also includes: Determine the quality inspection scope information of the current quality inspection cycle according to the detection scope of the current quality inspection cycle; Based on the quality-inspected range information of the current quality-inspection cycle, the cached quality-inspected range information is updated, and based on the single-sentence quality-inspection result of the current quality-inspection cycle, the cached historical quality-inspection result is updated.

8. The method according to claim 1, characterized in that Also includes: In response to the time series identifier of the incremental statement of the current quality inspection cycle being earlier than the time series identifier of the incremental statement of the previous quality inspection cycle, the quality inspection operation of the current quality inspection cycle is abandoned.

9. The method according to claim 1, characterized in that: After determining the sentence quality inspection result of the current quality inspection cycle, the method further includes: In response to the statement quality inspection result meeting a preset condition, a prompt message is generated.

10. A sentence quality inspection device, characterized in that: include: A global quality inspection module, configured to determine a global quality inspection result of a current quality inspection cycle based on a global quality inspection rule; A scope determination module is configured to determine the detection scope of the current quality inspection cycle from the full amount of sentences based on the single sentence quality inspection rules and the cached quality inspection scope information, wherein the full amount of sentences includes the incremental sentences of all quality inspection cycles; A single sentence quality inspection module is configured to perform single sentence quality inspection on the sentences within the detection range based on the single sentence quality inspection rule, and obtain the single sentence quality inspection result of the current quality inspection cycle in combination with the cached historical quality inspection results; The result determination module is configured to determine the sentence quality inspection result of the current quality inspection cycle according to the global quality inspection result and the single sentence quality inspection result.

11. A sentence quality inspection system, characterized in that: include: processor; as well as A memory storing computer instructions, wherein the computer instructions are used to cause the processor to execute the method according to any one of claims 1 to 9.

12. A storage medium, characterized in that: The storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 9.

Images