CN116822529B - Knowledge element extraction method based on semantic generalization - Google Patents
Knowledge element extraction method based on semantic generalization Download PDFInfo
- Publication number
- CN116822529B CN116822529B CN202311092677.8A CN202311092677A CN116822529B CN 116822529 B CN116822529 B CN 116822529B CN 202311092677 A CN202311092677 A CN 202311092677A CN 116822529 B CN116822529 B CN 116822529B
- Authority
- CN
- China
- Prior art keywords
- extraction
- generalization
- time
- voice
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000605 extraction Methods 0.000 title claims abstract description 128
- 238000011156 evaluation Methods 0.000 claims abstract description 61
- 238000012545 processing Methods 0.000 claims abstract description 43
- 238000000034 method Methods 0.000 claims abstract description 42
- 238000006243 chemical reaction Methods 0.000 claims abstract description 24
- 238000004891 communication Methods 0.000 claims abstract description 12
- 230000008569 process Effects 0.000 claims description 22
- 230000014509 gene expression Effects 0.000 claims description 16
- 238000007667 floating Methods 0.000 claims description 13
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000001105 regulatory effect Effects 0.000 claims description 8
- 230000002159 abnormal effect Effects 0.000 claims description 2
- 230000001276 controlling effect Effects 0.000 claims description 2
- 230000033228 biological regulation Effects 0.000 abstract description 8
- 238000004458 analytical method Methods 0.000 description 23
- 230000004044 response Effects 0.000 description 12
- 238000005516 engineering process Methods 0.000 description 6
- 230000003111 delayed effect Effects 0.000 description 4
- 238000003058 natural language processing Methods 0.000 description 4
- 230000007958 sleep Effects 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 230000000712 assembly Effects 0.000 description 1
- 238000000429 assembly Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013075 data extraction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000005059 dormancy Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 230000035484 reaction time Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
Landscapes
- Machine Translation (AREA)
Abstract
The invention discloses a knowledge element extraction method based on semantic generalization, and relates to the technical field of data processing; the method comprises the following steps: the method comprises the steps of collecting multiple data information of voice data, processing equipment operation information and communication conversion information to generate a generalization evaluation coefficient, comparing the generated generalization evaluation coefficient with a set generalization evaluation threshold value, evaluating the quality of received voice, determining the voice data needing to be subjected to generalization processing, analyzing according to data precision information and processing duration information, generating extraction influence factors, analyzing the extraction condition of knowledge elements, evaluating the extraction condition of the knowledge elements twice according to the comparison result of the extraction influence factors and the extraction evaluation threshold value, and performing corresponding operation according to different extraction conditions, so that the accuracy of intelligent home regulation is improved, and the high efficiency of intelligent home operation is ensured.
Description
Technical Field
The invention relates to the technical field of data extraction, in particular to a knowledge element extraction method based on semantic generalization.
Background
Semantic generalization refers to the process of converting specific and specific language expressions into more general and abstract semantic expressions in natural language processing, and by capturing potential semantic information of texts, processing equipment such as computers and the like can better understand and process diversified expression modes and text variants.
Knowledge elements refer to basic elements which have important significance in specific fields, have key roles in understanding and processing problems in related fields, and are a very important step in the tasks of information extraction, natural language processing, knowledge graph construction and the like.
The prior art has the following defects: along with development of science and technology, more and more intelligent home devices are applied to daily life and operate through voice of a user, but when the intelligent home devices receive voice, voice quality is too poor, so that semantic analysis is often unclear, regulation and control of the intelligent home devices cannot be performed correctly, wrong analysis results or overlong analysis time cause distrust of the intelligent home devices by the user, even the manufacturer is different, after the voice of the user is received, whether the voice needs semantic generalization processing cannot be rapidly judged, the situation of knowledge element extraction after the semantic generalization processing is unclear, and the operation cannot be regulated and controlled clearly, so that feedback time is overlong, inquiry resolution is unclear, and waste of time and resources is caused.
The above information disclosed in the background section is only for enhancement of understanding of the background of the disclosure and therefore it may include information that does not form the prior art that is already known to a person of ordinary skill in the art.
Disclosure of Invention
The invention aims to provide a knowledge element extraction method based on semantic generalization, which aims to solve the defects in the background technology.
In order to achieve the above object, the present invention provides the following technical solutions: the knowledge element extraction method based on semantic generalization comprises the following steps of;
collecting multiple data information of voice data, wherein the multiple data information comprises equipment operation information and communication conversion information;
generating a generalization evaluation coefficient by using equipment operation information and communication conversion information in the plurality of items of data information;
comparing the generated generalization evaluation coefficient with a generalization evaluation threshold value, and performing generalization processing on the received voice data according to the comparison result;
collecting element extraction information, wherein the element extraction information comprises data precision information and processing time length information, and analyzing the data precision information and the processing time length information to generate extraction influence factors;
and analyzing knowledge element extraction conditions of the generalized voice conversion text according to the extraction influence factors.
Preferably, the device operation information comprises wake-up influence time length and is calibrated asAllocating resource fluctuation range and calibrating to be +.>The communication conversion information includes delay signal-to-noise interference and is marked as +.>The data precision information comprises a call-in harmonic coefficient and is marked as +.>The processing duration information comprises the extraction duration floating coefficient and is marked as +.>。
Preferably, the logic for obtaining the wake-up influencing time length is as follows:
setting a plurality of time periods, and acquiring historical awakening of intelligent household equipment in the time periodsThe number HXi of times, namely obtaining a time difference value from each historical wake-up time to a corresponding sleep-entering time in a time period as a historical work operation time GZi, obtaining a time period set M= { M1, M2, … …, mi } contained in t time, wherein i is a positive integer, and calculating the historical work time and the historical wake-up number of the smart home in the time period set to obtain a wake-up influence time, wherein the calculation expression is as follows:where x represents the total number of time period set data.
Preferably, the logic for obtaining the fluctuation amplitude of the allocation resource is as follows:
acquiring the occupancy rates of a CPU and a memory of the intelligent home during voice recognition and coordinated control, wherein the occupancy rates of the voice recognition CPU and the memory are respectively as follows、/>The occupancy rates of the linkage control CPU and the memory are respectively +.>、/>Acquiring the CPU occupancy rate and the memory occupancy rate of each task in a time period T, calculating the square of the difference between the occupancy rate of each task and the average value, and carrying out summation processing to obtain a CPU resource occupancy value->Memory resource occupancy value->Calculating the standard deviation of the CPU resource occupation value and the standard deviation of the memory resource occupation value in the time period T, and summing the standard deviation of the CPU resource occupation value and the standard deviation of the memory resource occupation value to obtain the fluctuation range of the allocated resource ≡>。
Preferably, the logic for delaying the signal-to-noise interference level acquisition is as follows:
waveform data of a voice signal and background noise are obtained, the voice signal and the background noise are respectively marked as s (ts) and n (ts), wherein ts represents time, power of the voice signal and the background noise is obtained as Es and En respectively, and signal to noise ratio is calculatedThe time difference of the delay signal is obtained and marked as Et, the delay signal-to-noise interference degree is calculated, and the calculation expression is as follows: />。
Preferably, generating the generalization evaluation coefficient by using the device operation information and the communication conversion information in the multiple data information means that the generalization evaluation coefficient is generated by combining the wake-up influence duration, the distributed resource fluctuation range and the delayed signal-to-noise interference degree, wherein the wake-up influence duration, the distributed resource fluctuation range and the generalization evaluation coefficient are in direct proportion, and the delayed signal-to-noise interference degree and the generalization evaluation coefficient are in inverse proportion.
Preferably, the generated generalization evaluation coefficient is compared with a generalization evaluation threshold, and the specific process is as follows:
if the generalization evaluation coefficient is larger than the generalization evaluation threshold, judging that the voice quality is poor, and sending out a semantic generalization signal;
if the generalization evaluation coefficient is smaller than or equal to the generalization evaluation threshold, judging that the voice quality is excellent, and sending out a text interpretation signal.
Preferably, the obtaining logic of the standard harmonic coefficients is as follows:
the method comprises the steps of obtaining correctly extracted knowledge element data amount PZ, obtaining all extracted knowledge element data amount PA, and calculating to obtain extraction accuracy: zql=pz/PA, obtaining all the real knowledge element data quantity PX, and calculating to obtain the extraction recall: ZHL =px/PA, calculating the extraction accuracy and the extraction recall to obtain a recall harmonic coefficient, where the calculation expression is:；
the extraction duration floating coefficient acquisition logic is as follows:
acquiring and recording time of received voice, acquiring a time set Sc= { t1, t2, t3, … …, tn } spent for extracting the knowledge elements for the first time according to voice conversion text, acquiring a time set sf= { f1, f2, f3, … …, fn } spent for extracting the knowledge elements for the second time after generalization processing, wherein n is a positive integer, calculating average values after the time sets Sc and Sf are summarized, respectively calibrating the average values as Savg, calculating absolute values of differences between the average values and the two time sets as average value sets, marking the average values as sj= { S1, S2, S3, … …, sy } and y as positive integers, and calculating the expression as follows:wherein->And j is a positive integer for the total number of the data in the mean value set.
Preferably, the knowledge element extraction condition of the generalized voice conversion text is analyzed according to the extraction influence factors, and the specific steps are as follows:
generating extraction influence factors by calling the standard harmonic coefficients and the extraction duration floating coefficients;
comparing the extraction influence factor with an extraction evaluation threshold;
if the extraction influence factor is greater than or equal to the extraction evaluation threshold, judging that the extraction knowledge elements are abnormal, and sending an inquiry request to a user by the intelligent home;
if the extraction influence factor is smaller than the extraction evaluation threshold, judging that the extraction knowledge element is normal, and directly regulating and controlling the intelligent home.
In the technical scheme, the invention has the technical effects and advantages that:
the method carries out dimensionless treatment on the wake-up influence duration, the distribution resource fluctuation amplitude and the delay signal-to-noise interference degree, comprehensively analyzes and generates a generalization evaluation coefficient after removing units, compares the generated generalization evaluation coefficient with a set generalization evaluation threshold value, evaluates the quality of received voice, determines voice data needing to be subjected to generalization treatment, accurately distributes computational resources, carries out extraction of knowledge elements on a voice conversion text according to semantic generalization signals, the method comprises the steps of carrying out primary knowledge element extraction before voice generalization, carrying out secondary knowledge element extraction after semantic generalization, analyzing according to data precision information and processing time length information, generating extraction influence factors, comparing the extraction influence factors with an extraction evaluation threshold value, evaluating the extraction condition of the two knowledge elements according to a comparison result, and carrying out corresponding operation according to the evaluated extraction condition, so that the accuracy of intelligent home regulation is improved, and the high efficiency of intelligent home operation is ensured.
Drawings
For a clearer description of embodiments of the present application or of the solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments described in the present invention, and that other drawings may be obtained according to these drawings for a person skilled in the art.
FIG. 1 is a flow chart of a knowledge element extraction method based on semantic generalization of the present invention.
Detailed Description
Example embodiments will now be described more fully with reference to the accompanying drawings. However, the exemplary embodiments may be embodied in many forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of the example embodiments to those skilled in the art.
Example 1: the invention provides a knowledge element extraction method based on semantic generalization as shown in fig. 1, which comprises the following steps:
the method comprises the steps that user voice is taken as original text data, text contained in the user voice is disassembled, firstly, voice input of the user is required to be converted into text form, the step is called voice recognition, voice recognition technology converts voice signals of the user into text representation for subsequent processing, however, accents among different users have obvious differences, original voice of the user is taken as original voice sample data, voice characteristics of people are analyzed through voice print recognition technology to carry out identity verification or recognition, and the method can be used for confirming identity of a person, namely a unique user who emits the voice, and the method is a specific acquisition step;
collecting a voice sample of a user, wherein the user needs to provide a section of voice sample for training, typically reading out a specific short text or phrase, and using the sample for training a voiceprint recognition model;
from the collected voice sample, the voiceprint recognition system extracts a series of features which describe the voice characteristics of the user, wherein the common voiceprint features comprise mel frequency cepstrum coefficients and the like;
in the voiceprint recognition stage, the voiceprint recognition system compares a voice sample provided by a user with an established voiceprint model, the voiceprint recognition system analyzes the voice characteristics of the user and matches the voice characteristics with the model to judge whether the voice is matched with the voiceprint in the model, the voiceprint recognition system outputs a judging result according to the voiceprint recognition result, namely judging whether the voice is the voiceprint of a known user, if the voice is successfully matched, the voice source is confirmed to be an authorized user, if the voice source is failed to be matched, judging that the voice source is a strange user, and the intelligent household generally has wake-up words, so that the authorized user wakes up the household through the wake-up words;
in the process of collecting voice samples, due to the influence of factors such as distance, noise, speaking speed and the like, voice recognition quality is often low, accurate information cannot be obtained when the voice samples are converted into texts, and therefore intelligent home cannot clearly adjust modes;
after receiving a plurality of items of data in voice data, a receiving end of the intelligent home equipment establishes a quality prediction model, and manages the voice with poor receiving quality after analyzing the voice based on the quality prediction model;
collecting multiple data information of voice data, wherein the multiple data information comprises equipment operation information and communication conversion information;
plant operationThe information comprises wake-up influence time length, distributed resource fluctuation amplitude, the communication conversion information comprises delay signal-to-noise interference degree, and after acquisition, the wake-up influence time length, the distributed resource fluctuation amplitude and the delay signal-to-noise interference degree are respectively calibrated as、、/>；
The wake-up influence time length in the data information has an important influence on the quality of analysis received voice, wherein the wake-up influence time length is the duration time between when the intelligent household equipment is awakened and when the intelligent household equipment receives a relevant voice instruction, and the wake-up influence time length has the following influence on the intelligent household equipment:
unnecessary interference: the long-time awakening state can continuously trigger the intelligent home equipment under the condition of no need, so that unnecessary interference and misoperation are caused, the user experience is reduced, and the voice receiving is influenced;
system resource occupation: the continuous wake-up state can occupy system resources including a processor, a memory and the like, so that the system performance is reduced or the system is crashed, the voice receiving is influenced, the frequent wake-up and the dormancy can enable a circuit and the processor of the equipment to work frequently, a certain amount of heat is generated, the performance and the service life of the equipment can be influenced by long-time high temperature, and even the equipment can be damaged;
stability decreases: frequent wake-up and sleep operations may increase the complexity and workload of the device, resulting in reduced system stability, system crashes or anomalies;
therefore, the wake-up influence time length in a plurality of pieces of data information is obtained, and the voice quality received by the intelligent home can be analyzed;
the acquisition logic of the wake-up influence time length is as follows:
setting a plurality of time periods, acquiring the historical wake-up times HXi of the intelligent household equipment in the time periods,obtaining time difference values from each time of historical wake-up time to corresponding sleep entering in a time period as a historical work operation duration GZi, obtaining a time period set M= { M1, M2, … …, mi } contained in t time, wherein i is a positive integer, and calculating the historical work time and the historical wake-up times of the intelligent home in the time period set to obtain wake-up influence duration, wherein the calculation expression is as follows:wherein x represents the total number of time zone set data;
it should be noted that, the smart home device is generally provided with a sleep reaction time, when the smart home is awakened by the user through voice, the smart home enters a response state, the voice of the user is received at any moment, when the user is awakened, the next operation is not performed within a set response time, the smart home device enters the sleep state from the response state, so as to save energy consumption, and when the user is in the response state, the smart home resets the response state time after the user performs the next operation;
the fluctuation range of the distributed resources in the plurality of items of data information has important influence on the quality of analysis received voice, the fluctuation range of the distributed resources represents the fluctuation condition of the distributed computing power resources in the voice analysis process of the intelligent home equipment, and the excessive fluctuation range of the distributed resources can cause the following problems:
delay increases: the fluctuation of the computing power resources is large, so that the speed of processing voice data is unstable, when the distributed computing power resources are less, the processing time of voice analysis is possibly prolonged, thereby increasing the analysis delay and affecting the real-time response capability of intelligent household equipment to voice instructions;
the resolution error rate increases: the fluctuation of the computational power resources is large, so that the analysis algorithm can not fully utilize the resources to carry out accurate voice recognition, and when the allocated computational power is small, the situation of increasing the analysis error rate occurs, so that the instruction analysis is inaccurate;
system load instability: the fluctuation of the computational power resources is large, so that the system load is unstable, and when the distributed computational power resources are less, the system load is too heavy, and the stability and performance of the equipment are affected;
therefore, the fluctuation amplitude of the distributed resources is analyzed, and the state in the process of receiving the data by the intelligent home can be further analyzed;
the logic for obtaining the fluctuation amplitude of the allocation resource is as follows:
acquiring the occupancy rates of a CPU and a memory of the intelligent home during voice recognition and coordinated control, wherein the occupancy rates of the voice recognition CPU and the memory are respectively as follows、/>The occupancy rates of the linkage control CPU and the memory are respectively +.>、/>Acquiring the CPU occupancy rate and the memory occupancy rate of each task in a time period T, calculating the square of the difference between the occupancy rate of each task and the average value, and carrying out summation processing to obtain a CPU resource occupancy value->Memory resource occupancy value->Calculating the standard deviation of the CPU resource occupation value and the standard deviation of the memory resource occupation value in the time period T, and summing the standard deviation of the CPU resource occupation value and the standard deviation of the memory resource occupation value to obtain the fluctuation range of the allocated resource ≡>；
The CPU occupancy rate and the memory occupancy rate are collected according to the set frequency in the set time, and the specific set time is changed according to actual conditions.
The delay signal-to-noise interference level in the multiple data information has an important influence on the quality of the analysis received voice, the delay signal-to-noise interference level represents the time ratio between the effective voice signal and the background noise, namely the background noise possibly appears before or after the voice signal arrives at the equipment, and the following problems are more likely to appear as the delay signal-to-noise interference level is larger:
identifying the delay increase: delaying the signal-to-noise interference causes the device not to recognize immediately after receiving the speech signal, but to wait for a period of time to ensure that the complete speech signal can be captured, resulting in an increase in recognition delay, and a corresponding increase in time from user speaking to actual response of the device;
the false recognition rate increases: the delayed signal-to-noise interference causes the device to confuse the speech signal with background noise during the recognition process, increasing the possibility of false recognition, thereby causing erroneous operations to be performed;
system resource utilization decreases: in order to handle the delay signal-to-noise interference level, the device needs to buffer the voice signal for a period of time before recognition, thereby increasing the use of system resources and causing the reduction of the resource utilization rate;
there are two main manifestations of the delay signal-to-noise interference level, one is a pre-signal delay, and when the background noise occurs before the speech signal reaches the device, the pre-signal delay is called, which means that the device starts to receive the background noise before receiving the speech signal, and the speech signal may be interfered by the background noise; one is a post-signal delay, referred to as a post-signal delay when background noise occurs after the speech signal has arrived at the device, meaning that the device continues to receive background noise after the speech signal has been received, potentially affecting subsequent speech recognition and processing;
the logic for delaying the signal-to-noise-interference level acquisition is as follows:
waveform data of a voice signal and background noise are obtained, the voice signal and the background noise are respectively marked as s (ts) and n (ts), wherein ts represents time, power of the voice signal and the background noise is obtained as Es and En respectively, and signal to noise ratio is calculatedObtaining the time difference of the delay signal and marking as Et, calculating the delay signal-to-noiseInterference degree, calculation formula is->；
It should be noted that, the waveform data may collect the voice signal and the background noise through the microphone, or read the corresponding data from the audio file, in practical application, the signal and the noise are usually preprocessed, such as windowing, removing the dc component, etc., and the manner of obtaining the delay signal depends on the specific application, and may be obtained through hardware synchronization, time stamping, or other methods.
Comprehensively analyzing the voice received by the intelligent home according to the equipment operation information and the communication conversion information, and obtaining the quality of the received voice according to an analysis result;
dimensionless processing is carried out on the acquired wake-up influence time length, the acquired distributed resource fluctuation amplitude and the delay signal-to-noise interference degree, a generalization evaluation coefficient is generated after a unit is removed, and the generalization evaluation coefficient is calibrated as followsThe formula according to is:
in the method, in the process of the invention,、/>、/>respectively wake-up influence duration->Distribution resource fluctuation amplitude->Delayed signal-to-noise interference level>Is a preset proportionality coefficient of>、/>、/>Are all greater than 0;
the formula shows that the larger the wake-up influence time length is, the larger the fluctuation range of the allocated resources is, the smaller the delay signal-to-noise interference degree is, namely the generalization evaluation coefficientThe larger the representation value of (2) is, the worse the voice quality of the obtained user is, the more semantic errors are easy to occur in the voice analysis process, the smaller the wake-up influence duration is, the smaller the fluctuation range of the allocated resources is, the larger the delay signal-to-noise interference degree is, namely the generalization evaluation coefficient is->The smaller the representation value of (c) indicates the better the speech quality of the acquired user;
after the voice of the user is acquired, the possible receiving quality of the voice is poor due to various factors, when the voice is classified and judged, the voice quality is divided by setting a corresponding threshold standard, the voice meeting the threshold standard can be subjected to text transliteration by taking the threshold as a dividing line, the correct text containing instruction information is obtained, and the text which does not meet the threshold standard is subjected to semantic generalization processing, so that the analysis time is saved, and the semantic quality is improved;
comparing the generated generalization evaluation coefficient with a generalization evaluation threshold value, and judging whether to generalize the received voice according to the comparison result;
after the generated generalization evaluation coefficient is obtained, comparing the generated generalization evaluation coefficient with a generalization evaluation threshold, if the generalization evaluation coefficient is larger than the generalization evaluation threshold, sending out a semantic generalization signal to the voice, performing generalization processing to indicate that the quality of the received voice is poor, and performing semantic generalization processing after text conversion on the voice, wherein an explicit instruction is difficult to identify;
if the generalization evaluation coefficient is smaller than or equal to the generalization evaluation threshold, a text transliteration signal is sent to the voice, which indicates that the quality of the received voice is better, the requirement of a voice recognition instruction is met, and the text transliteration can be carried out on the voice;
the generalization process is to make the speech processing more fault tolerant and flexible, and to be able to recognize incomplete or ambiguous speech instructions, e.g. generalize specific device names, allowing the use of broader names or aliases to represent devices, or generalize specific operation instructions, allowing the use of hyponyms or related lexicons to represent operations.
According to the method, dimensionless processing is carried out on the wake-up influence duration, the fluctuation amplitude of the allocated resources and the delay signal-to-noise interference degree, the generalized evaluation coefficient is generated through comprehensive analysis after a unit is removed, the quality of received voice is evaluated according to comparison between the generated generalized evaluation coefficient and a set generalized evaluation threshold, so that voice data needing to be subjected to generalized processing is determined, and computational resources are accurately allocated.
Example 2: the method comprises the steps of extracting knowledge elements for the first time from a voice conversion text which sends out a semantic generalization signal, and identifying the knowledge elements in the voice conversion text as initial knowledge elements;
word segmentation is carried out on a text (voice conversion text), a text sentence is segmented into a series of meaningful words or phrases to form semantic units, a certain error tolerance mechanism is introduced for the recognized fuzzy text, a certain degree of error or fuzzy matching is allowed, for example, an editing distance-based method is used for carrying out approximate matching on the recognized text and words in a predefined dictionary, and the most similar words are found to be used as word segmentation results;
part of speech tagging is performed on the word or phrase after word segmentation, namely, the part of speech (noun, verb, adjective, etc.) of each word is judged, so as to help understand the grammar structure and semantic meaning of sentences;
identifying specific entities in the text, such as name, place name, date, time and other information, by using named entity identification (NER) and other technologies, and identifying the entities such as equipment name, place, time and the like in the intelligent home;
semantic understanding is carried out on text or voice by using a Natural Language Processing (NLP) technology, the text or voice is converted into semantic representation which can be understood by a computer, and the steps involve tasks such as grammar analysis, syntax structure analysis, semantic role marking and the like;
based on semantic understanding, specific knowledge element extraction is carried out according to the function of the intelligent home, for example, information such as equipment name, operation instruction, time and the like is extracted from semantic representation;
after the first knowledge element extraction, performing generalization processing on the extracted knowledge element, wherein the generalization processing is to enable the system to have higher fault tolerance and flexibility, and can identify incomplete voice instructions or voice instructions containing fuzzy information, for example, generalizing specific equipment names, allowing the equipment to be represented by using wider names or aliases, or generalizing specific operation instructions, allowing the operation to be represented by using hyponyms or related word assemblies;
after the text is subjected to the generalization processing, the text subjected to the generalization processing is subjected to second extraction of knowledge elements again, the knowledge elements subjected to the generalization processing are identified, and as the second knowledge elements, the second extraction of knowledge elements can be regarded as a process of carrying out semantic understanding on the text subjected to the generalization again so as to acquire more specific and accurate knowledge elements;
the first knowledge element extraction and the second knowledge element extraction have a close relation, key information in an initial voice instruction is obtained through the first extraction, then generalization processing is carried out, the obtained generalization text possibly contains some fuzzy or incomplete information, the second extraction carries out deeper semantic understanding on the generalization text, more specific knowledge elements are obtained, and through the two extraction and generalization processes, the intelligent home system can better understand the intention of a user and respond;
collecting element extraction information including a numberAccording to the precision information and the processing time length information, the data precision information comprises a calling regulating coefficient, the processing time length information comprises an extraction time length floating coefficient, and after acquisition, the calling regulating coefficient and the extraction time length floating coefficient are respectively calibrated as、/>；
The recall-in reconciliation coefficient plays an important role in analyzing the extraction condition of the knowledge elements, and represents the condition between the extraction accuracy and the recall rate in the process of extracting the knowledge elements, and can be used for comprehensively evaluating the performance of the model, and the Gao Zhaozhun reconciliation coefficient means that the model is good in accuracy and recall rate, can accurately extract the knowledge elements and covers more real knowledge elements;
the acquisition logic of the recall harmonic coefficients is as follows:
the method comprises the steps of obtaining correctly extracted knowledge element data amount PZ, obtaining all extracted knowledge element data amount PA, and calculating to obtain extraction accuracy: zql=pz/PA, obtaining all the real knowledge element data quantity PX, and calculating to obtain the extraction recall: ZHL =px/PA, calculating the extraction accuracy and the extraction recall to obtain a recall harmonic coefficient, where the calculated expression is:；
the extraction duration floating coefficient plays an important role in the extraction condition of the analysis knowledge elements, and has the following effects:
the response speed becomes slow: the extraction duration of the knowledge elements directly influences the response speed of the intelligent home to the voice instruction, if the extraction duration is longer, a user may need to wait for a longer time to obtain the response of the system, and the user experience and the interaction instantaneity are reduced;
real-time performance becomes poor: the intelligent home needs to process voice instructions in real time and quickly make corresponding operations, if the extraction duration of the knowledge elements is too long, the system cannot respond to the instructions of the user in real time, and the real-time performance of the intelligent home is affected;
system load imbalance: the extraction of the knowledge elements may consume more computing resources and computing power, and if the extraction time is longer, the load of the system is increased, and the execution efficiency of other tasks is affected;
the extraction duration floating coefficient acquisition logic is as follows:
acquiring and recording time of received voice, acquiring a time set Sc= { t1, t2, t3, … …, tn } spent according to the first extraction of the knowledge elements of the text, wherein n is a positive integer, acquiring a time set sf= { f1, f2, f3, … …, fn } spent for the second extraction of the knowledge elements after generalization, wherein n is a positive integer, calculating an average value obtained by integrating the time sets Sc and Sf, respectively calibrating the average value as Savg, calculating an absolute value of a difference value between the average value and the two time sets as an average value set, marking the average value as sj= { S1, S2, S3, … …, sy } and y as the positive integer, and calculating the expression as follows:wherein->The total data quantity is the average value set, j is a positive integer;
dimensionless processing is carried out on the obtained call criterion harmonic coefficient and the extraction duration floating coefficient, extraction influence factors are generated after units are removed, and the extraction influence factors are calibrated as followsThe formula according to is:
in the method, in the process of the invention,、/>respectively called standard regulating coefficient->Extraction duration floating coefficient +.>Is a preset proportionality coefficient of>、/>Are all greater than 0;
as can be seen from the formula, the smaller the calling harmonic coefficient is, the larger the extraction duration floating coefficient is, namely the extraction influence factorThe larger the expression value of (2) shows that the worse the extraction efficiency of knowledge elements in the acquired voice is, the requirement of operation instantaneity is not met, the larger the calling harmonic coefficient is, the smaller the extraction duration floating coefficient is, namely the extraction influence factor->The smaller the expression value of (2) is, the better the extraction efficiency of knowledge elements in the acquired voice is, and the higher the accuracy is when the voice instruction is executed;
it should be noted that, the integrated performance monitoring tool is provided in the smart home, which can monitor performance of the system in real time, including indexes such as extraction duration and response speed, real-time data of the extraction duration can be obtained through the performance monitoring tool, the log is recorded in the smart home system, and the receiving time and the extraction completion time of the voice command are also included, the extraction duration of each voice command can be obtained through analyzing the log data, and the time from the start of receiving voice to the time of extracting each knowledge element is counted one by one, for example, the extraction time of two knowledge elements is 3.2s,3.3s, the time of receiving voice is indicated as the starting time, it takes 3.2s to extract the first knowledge element, and it takes 3.3s to extract the second knowledge element;
comparing the extraction influence factor with an extraction evaluation threshold;
if the extraction influence factor is greater than or equal to the extraction evaluation threshold value, indicating that the condition of extracting the knowledge elements is poor, and the knowledge elements cannot be correspondingly regulated and controlled according to the existing analysis, and the intelligent home sends an inquiry request to the user;
if the extraction influence factor is smaller than the extraction evaluation threshold, the situation of extracting the knowledge elements is better, the regulation and control conditions are met, and the intelligent home is regulated and controlled;
the intelligent household equipment sends an inquiry request to a user, the user needs clear second voice instruction transmission, after the user carries out the second voice instruction transmission, the intelligent household equipment carries out corresponding regulation and control according to the second voice, if the intelligent household equipment can carry out successful regulation and control according to the second voice instruction, the first voice data and the second voice data are put into a related model for analysis, after similar conditions are met, more intelligent analysis and control can be carried out, if corresponding regulation and control information cannot not be obtained, the intelligent household equipment is warned, and the intelligent household equipment possibly has problems;
it should be noted that, in this embodiment, the relevant threshold information is preset by a professional, for example, the extraction evaluation threshold is set according to the recognition speed, response time and recognition accuracy of the knowledge element as criteria, and other threshold setting manners are not explained here too much.
According to the method and the device, knowledge elements are extracted from the voice conversion text according to the semantic generalization signal, the knowledge elements are extracted once before voice generalization, the knowledge elements are extracted for the second time after semantic generalization, analysis is carried out according to the data precision information and the processing time length information, extraction influence factors are generated, the extraction influence factors are compared with the extraction evaluation threshold value, the extraction conditions of the knowledge elements are evaluated twice according to the comparison result, corresponding operation is carried out according to the evaluated extraction conditions, and therefore the accuracy of intelligent home regulation is improved, and the high efficiency of intelligent home operation is guaranteed.
The above formulas are all formulas with dimensions removed and numerical values calculated, the formulas are formulas with a large amount of data collected for software simulation to obtain the latest real situation, and preset parameters in the formulas are set by those skilled in the art according to the actual situation.
While certain exemplary embodiments of the present invention have been described above by way of illustration only, it will be apparent to those of ordinary skill in the art that modifications may be made to the described embodiments in various different ways without departing from the spirit and scope of the invention. Accordingly, the drawings and description are to be regarded as illustrative in nature and not as restrictive of the scope of the invention, which is defined by the appended claims.
It is noted that relational terms such as first and second, and the like, if any, are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises an element.
It should be understood that, in various embodiments of the present application, the sequence numbers of the foregoing processes do not mean the order of execution, and the order of execution of the processes should be determined by the functions and internal logic thereof, and should not constitute any limitation on the implementation process of the embodiments of the present application.
The foregoing is merely specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily think about changes or substitutions within the technical scope of the present application, and the changes and substitutions are intended to be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (4)
1. The knowledge element extraction method based on semantic generalization is characterized by comprising the following steps of;
collecting multiple data information of voice data, wherein the multiple data information comprises equipment operation information and communication conversion information;
generating a generalization evaluation coefficient by using equipment operation information and communication conversion information in the plurality of items of data information;
comparing the generated generalization evaluation coefficient with a generalization evaluation threshold value, and performing generalization processing on the received voice data according to the comparison result;
collecting element extraction information, wherein the element extraction information comprises data precision information and processing time length information, and analyzing the data precision information and the processing time length information to generate extraction influence factors;
analyzing knowledge element extraction conditions of the generalization processed voice conversion text according to the extraction influence factors;
the equipment operation information comprises the wake-up influence time length and is calibrated asAllocating resource fluctuation range and calibrating to be +.>The communication conversion information includes delay signal-to-noise interference and is marked as +.>The data precision information comprises a call-in harmonic coefficient and is marked as +.>The processing duration information comprises the extraction duration floating coefficient and is marked as +.>；
Setting a plurality of time periods, acquiring historical wake-up times HXi of intelligent household equipment in the time periods, and acquiring each time periodThe time difference from the secondary historical wake-up time to the corresponding sleep-entering time is taken as a historical work operation duration GZi, a time period set M= { M1, M2, … …, mi } contained in the t time is obtained, i is a positive integer, the historical work time and the historical wake-up times of the intelligent home in the time period set are calculated to obtain wake-up influence duration, and a calculation expression is as follows:wherein x represents the total number of time zone set data;
acquiring the occupancy rates of a CPU and a memory of the intelligent home during voice recognition and coordinated control, wherein the occupancy rates of the voice recognition CPU and the memory are respectively as follows、/>The occupancy rates of the linkage control CPU and the memory are respectively +.>、/>Acquiring the CPU occupancy rate and the memory occupancy rate of each task in a time period T, calculating the square of the difference between the occupancy rate of each task and the average value, and carrying out summation processing to obtain a CPU resource occupancy value->Memory resource occupancy value->Calculating the standard deviation of the CPU resource occupation value and the standard deviation of the memory resource occupation value in the time period T, and summing the standard deviation of the CPU resource occupation value and the standard deviation of the memory resource occupation value to obtain the fluctuation range of the allocated resource ≡>；
Waveform data of a voice signal and background noise are obtained, the voice signal and the background noise are respectively marked as s (ts) and n (ts), wherein ts represents time, power of the voice signal and the background noise is obtained as Es and En respectively, and signal to noise ratio is calculatedThe time difference of the delay signal is obtained and marked as Et, the delay signal-to-noise interference degree is calculated, and the calculation expression is as follows: />；
The method comprises the steps of obtaining correctly extracted knowledge element data amount PZ, obtaining all extracted knowledge element data amount PA, and calculating to obtain extraction accuracy: zql=pz/PA, obtaining all the real knowledge element data quantity PX, and calculating to obtain the extraction recall: ZHL =px/PA, calculating the extraction accuracy and the extraction recall to obtain a recall harmonic coefficient, where the calculation expression is:；
acquiring and recording time of received voice, acquiring a time set Sc= { t1, t2, t3, … …, tn } spent for extracting the knowledge elements for the first time according to voice conversion text, acquiring a time set sf= { f1, f2, f3, … …, fn } spent for extracting the knowledge elements for the second time after generalization processing, wherein n is a positive integer, calculating average values after the time sets Sc and Sf are summarized, respectively calibrating the average values as Savg, calculating absolute values of differences between the average values and the two time sets as average value sets, marking the average values as sj= { S1, S2, S3, … …, sy } and y as positive integers, and calculating the expression as follows:wherein->And j is a positive integer for the total number of the data in the mean value set.
2. The semantic generalization-based knowledge element extraction method according to claim 1, wherein generating a generalization evaluation coefficient from device operation information and communication conversion information in a plurality of data information means generating a generalization evaluation coefficient by combining a wake-up influence duration, an allocation resource fluctuation range and a delay signal-to-noise interference degree, wherein the wake-up influence duration, the allocation resource fluctuation range and the generalization evaluation coefficient are in direct proportion, and the delay signal-to-noise interference degree is in inverse proportion to the generalization evaluation coefficient.
3. The semantic generalization-based knowledge element extraction method according to claim 2, characterized in that the generated generalization evaluation coefficient is compared with a generalization evaluation threshold, and the specific process is as follows:
if the generalization evaluation coefficient is larger than the generalization evaluation threshold, judging that the voice quality is poor, and sending out a semantic generalization signal;
if the generalization evaluation coefficient is smaller than or equal to the generalization evaluation threshold, judging that the voice quality is excellent, and sending out a text interpretation signal.
4. The semantic generalization-based knowledge element extraction method according to claim 3, wherein the knowledge element extraction condition of the generalization processed voice conversion text is analyzed according to the extraction influence factor, and the specific steps are as follows:
generating extraction influence factors by calling the standard harmonic coefficients and the extraction duration floating coefficients;
comparing the extraction influence factor with an extraction evaluation threshold;
if the extraction influence factor is greater than or equal to the extraction evaluation threshold, judging that the extraction knowledge elements are abnormal, and sending an inquiry request to a user by the intelligent home;
if the extraction influence factor is smaller than the extraction evaluation threshold, judging that the extraction knowledge element is normal, and directly regulating and controlling the intelligent home.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311092677.8A CN116822529B (en) | 2023-08-29 | 2023-08-29 | Knowledge element extraction method based on semantic generalization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311092677.8A CN116822529B (en) | 2023-08-29 | 2023-08-29 | Knowledge element extraction method based on semantic generalization |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116822529A CN116822529A (en) | 2023-09-29 |
| CN116822529B true CN116822529B (en) | 2023-12-29 |
Family
ID=88115288
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311092677.8A Active CN116822529B (en) | 2023-08-29 | 2023-08-29 | Knowledge element extraction method based on semantic generalization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116822529B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116386617A (en) * | 2023-03-16 | 2023-07-04 | 淮南职业技术学院 | English semantic recognition analysis method based on big data |
| CN116644756A (en) * | 2023-05-04 | 2023-08-25 | 北京海致科技集团有限公司 | Semantic Analysis Method Based on Knowledge Graph |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107360327B (en) * | 2017-07-19 | 2021-05-07 | 腾讯科技（深圳）有限公司 | Speech recognition method, apparatus and storage medium |
-
2023
- 2023-08-29 CN CN202311092677.8A patent/CN116822529B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116386617A (en) * | 2023-03-16 | 2023-07-04 | 淮南职业技术学院 | English semantic recognition analysis method based on big data |
| CN116644756A (en) * | 2023-05-04 | 2023-08-25 | 北京海致科技集团有限公司 | Semantic Analysis Method Based on Knowledge Graph |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116822529A (en) | 2023-09-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105679310A (en) | Method and system for speech recognition | |
| CN110047481B (en) | Method and apparatus for speech recognition | |
| US7177810B2 (en) | Method and apparatus for performing prosody-based endpointing of a speech signal | |
| CN105550361B (en) | Log processing method and device and question and answer information processing method and device | |
| CN111081218A (en) | Voice recognition method and voice control system | |
| CN114974229A (en) | Method and system for extracting abnormal behaviors based on audio data of power field operation | |
| WO2022222045A1 (en) | Speech information processing method, and device | |
| CN116822529B (en) | Knowledge element extraction method based on semantic generalization | |
| CN110889008B (en) | Music recommendation method and device, computing device and storage medium | |
| CN112087726B (en) | Method and system for identifying polyphonic ringtone, electronic equipment and storage medium | |
| CN112309372B (en) | Intent recognition method, device, equipment and storage medium based on intonation | |
| CN111862943A (en) | Speech recognition method and apparatus, electronic device, and storage medium | |
| CN113763962A (en) | Audio processing method and device, storage medium and computer equipment | |
| CN110910905B (en) | Mute point detection method and device, storage medium and electronic equipment | |
| CN113241063B (en) | Algorithm parameter updating method, device, terminal and medium in voice recognition system | |
| CN111176618B (en) | Method and system for developing program by voice wakeup | |
| CN113643700A (en) | Control method and system of intelligent voice switch | |
| CN114420103A (en) | Voice processing method and device, electronic equipment and storage medium | |
| CN114707515A (en) | Method and device for judging dialect, electronic equipment and storage medium | |
| CN114254628A (en) | Method and device for quickly extracting hot words by combining user text in voice transcription, electronic equipment and storage medium | |
| CN109446527B (en) | Nonsensical corpus analysis method and system | |
| CN113421572B (en) | Real-time audio dialogue report generation method and device, electronic equipment and storage medium | |
| CN111354365A (en) | Pure voice data sampling rate identification method, device and system | |
| Girirajan et al. | Hybrid Feature Extraction Technique for Tamil Automatic Speech Recognition System in Noisy Environment | |
| CN117894321B (en) | Voice interaction method, voice interaction prompting system and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |