CN109033082A - The learning training method, apparatus and computer readable storage medium of semantic model - Google Patents
The learning training method, apparatus and computer readable storage medium of semantic model Download PDFInfo
- Publication number
- CN109033082A CN109033082A CN201810800318.6A CN201810800318A CN109033082A CN 109033082 A CN109033082 A CN 109033082A CN 201810800318 A CN201810800318 A CN 201810800318A CN 109033082 A CN109033082 A CN 109033082A
- Authority
- CN
- China
- Prior art keywords
- word
- segmentation result
- word segmentation
- total amount
- semantic model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a kind of learning training methods of semantic model, it include: based on preset core word bank, word segmentation processing is carried out to training text using two-way maximum matching method, obtains the corresponding matching word segmentation result of the training text, the matching word segmentation result includes individual character and/or core word;When there are core word, the corresponding dictionary type of the core word is determined;According to preset semantic model, the individual character and/or the corresponding index coding of the dictionary type are obtained, with the training text according to the index coded representation, wherein, in the semantic model, the corresponding unique index coding of each individual character, the corresponding unique index coding of each dictionary type.The invention also discloses the learning training devices and computer readable storage medium of a kind of semantic model.The present invention improves the learning training efficiency of semantic model.
Description
Technical field
The present invention relates to field of computer technology more particularly to the learning training method, apparatus and meter of a kind of semantic model
Calculation machine readable storage medium storing program for executing.
Background technique
Semantic model is to increase completely new data builder and data processing primitive on the basis of relational model, is used to table
Up to a new class of data model of complicated structure and semanteme abundant, currently, being usually by each word using unique index
Encoding index indicates, since the order of magnitude of dictionary is often in hundreds of thousands grade, then the order of magnitude of corresponding index coding is also several
100000 grades, this will lead to dimension disaster, and semantic model is especially big, and learning training speed is especially slow, low efficiency.
Summary of the invention
The main object of the present invention is to provide the learning training method, apparatus and computer-readable storage of a kind of semantic model
Medium, it is intended to solve the problems, such as the learning training low efficiency of existing semantic model.
To achieve the above object, the present invention proposes the learning training method of semantic model, the study instruction of the semantic model
Practice method the following steps are included:
Based on preset core word bank, word segmentation processing is carried out to training text using two-way maximum matching method, described in acquisition
The corresponding matching word segmentation result of training text, the matching word segmentation result includes individual character and/or core word;
When there are core word, the corresponding dictionary type of the core word is determined;
According to preset semantic model, the individual character and/or the corresponding index coding of the dictionary type are obtained, with basis
Training text described in the index coded representation, wherein in the semantic model, the corresponding unique index of each individual character is compiled
Code, the corresponding unique index coding of each dictionary type.
Preferably, described to be based on preset core word bank, training text is carried out at participle using two-way maximum matching method
Reason, the step of obtaining the training text corresponding matching word segmentation result include:
Word segmentation processing is carried out to training text using two-way maximum matching method, obtains the corresponding individual character of the training text
And/or word;
When there are word, it is based on the core word bank, judges whether institute's predicate is core word;
If it is not, institute's predicate is then split as individual character, the corresponding matching word segmentation result of the training text is obtained.
Preferably, described when there are word, it is based on the core word bank, judges the step of whether institute's predicate is core word packet
It includes:
If the quantity of institute's predicate be it is multiple, be based on the core word bank, judge whether each word is core word respectively;
It is described if it is not, institute's predicate is then split as individual character, obtain the step of the corresponding matching word segmentation result of the training text
Suddenly include:
Word is not core word if it exists, then institute's predicate is split as individual character, obtains the corresponding matching point of the training text
Word result.
Preferably, described that word segmentation processing is carried out to training text using two-way maximum matching method, obtain the training text
The step of corresponding individual character and/or word includes:
Word segmentation processing is carried out to training text using two-way maximum matching method, it is two-way point corresponding to obtain the training text
Word result;
According to the word frequency data counted in advance, word frequency analysis is carried out to the two-way word segmentation result, determines the training text
Corresponding individual character and/or word.
Preferably, the two-way word segmentation result includes that positive word segmentation result and reverse word segmentation result, the basis count in advance
Word frequency data, word frequency analysis is carried out to the two-way word segmentation result, determines the corresponding individual character of the training text and/or word
Step includes:
According to the word frequency data, word frequency analysis is carried out to the positive word segmentation result and reverse word segmentation result respectively, is obtained
Take the positive corresponding first word frequency total amount of word segmentation result and the corresponding second word frequency total amount of the reverse word segmentation result;
According to the first word frequency total amount and the second word frequency total amount, the positive word segmentation result or described reverse is determined
Word segmentation result is the matching word segmentation result;
By the matching word segmentation result individual character and/or word that include, be determined as the corresponding individual character of the training text and/or
Word.
Preferably, described according to the first word frequency total amount and the second word frequency total amount, determine the positive participle knot
The step of fruit or the reverse word segmentation result are the matching word segmentation result include:
When the ratio of the first word frequency total amount and the second word frequency total amount is greater than default ratio, the forward direction is determined
Word segmentation result is the matching word segmentation result;
When the ratio of the first word frequency total amount and the second word frequency total amount is less than the default ratio, described in determination
Reverse word segmentation result is the matching word segmentation result.
Preferably, described according to the first word frequency total amount and the second word frequency total amount, determine the positive participle knot
The step of fruit or reverse word segmentation result are the matching word segmentation result include:
When the difference of the first word frequency total amount and the second word frequency total amount is greater than preset difference value, the forward direction is determined
Word segmentation result is the matching word segmentation result;
When the difference of the first word frequency total amount and the second word frequency total amount is less than the preset difference value, described in determination
Reverse word segmentation result is the matching word segmentation result.
Preferably, described according to the word frequency data, the positive word segmentation result and reverse word segmentation result are carried out respectively
Word frequency analysis obtains the positive corresponding first word frequency total amount of word segmentation result and the reverse word segmentation result corresponding the
The step of two word frequency total amounts includes:
According to the word frequency data, individual character and/or the corresponding word frequency amount of word that the positive word segmentation result includes are carried out
It is cumulative, obtain the corresponding first word frequency total amount of the positive word segmentation result;
According to the word frequency data, the corresponding word frequency amount of the individual character and/or word for including by the reverse word segmentation result is carried out
It is cumulative, obtain the corresponding second word frequency total amount of the reverse word segmentation result.
In addition, to achieve the above object, the present invention also proposes a kind of learning training device of semantic model, the semanteme mould
The learning training device of type includes: memory, processor and is stored on the memory and can run on the processor
Semantic model learning training program;It realizes when the semantic model learning training program is executed by the processor such as institute above
The step of learning training method for the semantic model stated.
In addition, to achieve the above object, it is described computer-readable the present invention also provides a kind of computer readable storage medium
Semantic model learning training program is stored on storage medium, it is real when the semantic model learning training program is executed by processor
The step of learning training method of semantic model now as described above.
In technical solution of the present invention, when carrying out semantic model learning training, it is based on preset core word bank, use is two-way
Maximum matching method carries out word segmentation processing to training text, obtain the corresponding matching word segmentation result of training text (including individual character and/or
Core word), when there are core word, determine the corresponding dictionary type of core word, and according to preset semantic model, obtain individual character
And/or dictionary type corresponding index coding, with according to acquired index coded representation training text, due in semantic model
In each individual character, dictionary type respectively correspond unique index coding, rather than the corresponding unique index coding of each word, individual character
Quantity be far smaller than the quantity of word, the order of magnitude of semantic model reduces, and this improves the learning training of semantic model effects
Rate.
Detailed description of the invention
Fig. 1 is that the learning training apparatus structure of the semantic model for the hardware running environment that the embodiment of the present invention is related to shows
It is intended to;
Fig. 2 is the flow diagram of the learning training method first embodiment of semantic model of the invention;
Fig. 3 is the flow diagram of the learning training method second embodiment of semantic model of the invention.
The embodiments will be further described with reference to the accompanying drawings for the realization, the function and the advantages of the object of the present invention.
Specific embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
The solution of the embodiment of the present invention is mainly: when carrying out semantic model learning training, being based on preset core
Dictionary carries out word segmentation processing to training text using two-way maximum matching method, obtains the corresponding matching word segmentation result of training text
(including individual character and/or core word) determines the corresponding dictionary type of core word, and according to preset language when there are core word
Adopted model obtains individual character and/or the corresponding index coding of dictionary type, according to acquired index coded representation training text
This, since individual character each in semantic model, dictionary type respectively correspond unique index coding, rather than each word is corresponding only
One index coding, the quantity of individual character are far smaller than the quantity of word, and the order of magnitude of semantic model substantially reduces, therefore, greatly
Improve the learning training efficiency of semantic model.Technical solution through the embodiment of the present invention solves the study of semantic model
The low problem of training effectiveness.
The embodiment of the present invention proposes a kind of learning training device of semantic model.
Referring to Fig.1, Fig. 1 is the learning training dress of the semantic model for the hardware running environment that the embodiment of the present invention is related to
Set structural schematic diagram.
In subsequent description, it is only using the suffix for indicating such as " module ", " component " or " unit " of element
Be conducive to explanation of the invention, itself there is no a specific meaning.Therefore, " module ", " component " or " unit " can mix
Ground uses.
As shown in Figure 1, the learning training device of the semantic model may include: processor 1001, communication bus 1002, use
Family interface 1003, network interface 1004, memory 1005.Wherein, communication bus 1002 is for realizing the company between these components
Connect letter.User interface 1003 may include display screen (Display), input unit such as keyboard (Keyboard), can be selected
Family interface 1003 can also include standard wireline interface and wireless interface.Network interface 1004 optionally may include standard
Wireline interface, wireless interface (such as WI-FI interface).Memory 1005 can be high speed RAM memory, be also possible to stable deposit
Reservoir (non-volatile memory), such as magnetic disk storage.Memory 1005 optionally can also be independently of aforementioned place
Manage the storage device of device 1001.
It will be understood by those skilled in the art that the learning training apparatus structure of semantic model shown in Fig. 1 is not constituted
Restriction to the learning training device of semantic model may include than illustrating more or fewer components, or the certain portions of combination
Part or different component layouts.
As shown in Figure 1, as may include that operating system, network are logical in a kind of memory 1005 of computer storage medium
Believe module and semantic model learning training program.
In the present invention, the learning training device of the semantic model is called in memory 1005 by processor 1001 to be stored
Semantic model learning training program, and execute following operation:
Based on preset core word bank, word segmentation processing is carried out to training text using two-way maximum matching method, described in acquisition
The corresponding matching word segmentation result of training text, the matching word segmentation result includes individual character and/or core word;
When there are core word, the corresponding dictionary type of the core word is determined;
According to preset semantic model, the individual character and/or the corresponding index coding of the dictionary type are obtained, with basis
Training text described in the index coded representation, wherein in the semantic model, the corresponding unique index of each individual character is compiled
Code, the corresponding unique index coding of each dictionary type.
Further, processor 1001 can call the semantic model learning training program stored in memory 1005, also
Execute following operation:
Word segmentation processing is carried out to training text using two-way maximum matching method, obtains the corresponding individual character of the training text
And/or word;
When there are word, it is based on the core word bank, judges whether institute's predicate is core word;
If it is not, institute's predicate is then split as individual character, the corresponding matching word segmentation result of the training text is obtained.
Further, processor 1001 can call the semantic model learning training program stored in memory 1005, also
Execute following operation:
If the quantity of institute's predicate be it is multiple, be based on the core word bank, judge whether each word is core word respectively;
It is described if it is not, institute's predicate is then split as individual character, obtain the step of the corresponding matching word segmentation result of the training text
Suddenly include:
Word is not core word if it exists, then institute's predicate is split as individual character, obtains the corresponding matching point of the training text
Word result.
Further, processor 1001 can call the semantic model learning training program stored in memory 1005, also
Execute following operation:
Word segmentation processing is carried out to training text using two-way maximum matching method, it is two-way point corresponding to obtain the training text
Word result;
According to the word frequency data counted in advance, word frequency analysis is carried out to the two-way word segmentation result, determines the training text
Corresponding individual character and/or word.
Further, processor 1001 can call the semantic model learning training program stored in memory 1005, also
Execute following operation:
According to the word frequency data, word frequency analysis is carried out to the positive word segmentation result and reverse word segmentation result respectively, is obtained
Take the positive corresponding first word frequency total amount of word segmentation result and the corresponding second word frequency total amount of the reverse word segmentation result;
According to the first word frequency total amount and the second word frequency total amount, the positive word segmentation result or described reverse is determined
Word segmentation result is the matching word segmentation result;
By the matching word segmentation result individual character and/or word that include, be determined as the corresponding individual character of the training text and/or
Word breath.
Further, processor 1001 can call the semantic model learning training program stored in memory 1005, also
Execute following operation:
When the ratio of the first word frequency total amount and the second word frequency total amount is greater than default ratio, the forward direction is determined
Word segmentation result is the matching word segmentation result;
When the ratio of the first word frequency total amount and the second word frequency total amount is less than the default ratio, described in determination
Reverse word segmentation result is the matching word segmentation result.
Further, processor 1001 can call the semantic model learning training program stored in memory 1005, also
Execute following operation:
When the difference of the first word frequency total amount and the second word frequency total amount is greater than preset difference value, the forward direction is determined
Word segmentation result is the matching word segmentation result;
When the difference of the first word frequency total amount and the second word frequency total amount is less than the preset difference value, described in determination
Reverse word segmentation result is the matching word segmentation result.
Further, processor 1001 can call the semantic model learning training program stored in memory 1005, also
Execute following operation:
According to the word frequency data, individual character and/or the corresponding word frequency amount of word that the positive word segmentation result includes are carried out
It is cumulative, obtain the corresponding first word frequency total amount of the positive word segmentation result;
According to the word frequency data, the corresponding word frequency amount of the individual character and/or word for including by the reverse word segmentation result is carried out
It is cumulative, obtain the corresponding second word frequency total amount of the reverse word segmentation result.
The present embodiment through the above scheme, when carrying out semantic model learning training, is based on preset core word bank, uses
Two-way maximum matching method carries out word segmentation processing to training text, obtains the corresponding matching word segmentation result of training text (including individual character
And/or core word), when there are core word, determine the corresponding dictionary type of core word, and according to preset semantic model, obtain
Individual character and/or dictionary type corresponding index coding are taken, with according to acquired index coded representation training text, due in language
Each individual character, dictionary type respectively correspond unique index coding in adopted model, rather than the corresponding unique index of each word is compiled
Code, the quantity of individual character are far smaller than the quantity of word, and the order of magnitude of semantic model substantially reduces, and therefore, greatly improves semanteme
The learning training efficiency of model.
Based on above-mentioned hardware configuration, the learning training embodiment of the method for semantic model of the present invention is proposed.
It is the flow diagram of the learning training method first embodiment of semantic model of the present invention referring to Fig. 2, Fig. 2.
In the first embodiment, the semantic model learning training method the following steps are included:
Step S10 is based on preset core word bank, carries out word segmentation processing to training text using two-way maximum matching method,
The corresponding matching word segmentation result of the training text is obtained, the matching word segmentation result includes individual character and/or core word;
Step S20 determines the corresponding dictionary type of the core word when there are core word;
Step S30 obtains the individual character and/or the corresponding index of the dictionary type is compiled according to preset semantic model
Code, with the training text according to the index coded representation, wherein in the semantic model, each individual character is corresponding unique
Index coding, the corresponding unique index coding of each dictionary type.
Currently, being usually by each word using unique index coded representation, since the order of magnitude of dictionary is often tens
Ten thousand grades, then the order of magnitude of corresponding index coding is also hundreds of thousands grade, this will lead to dimension disaster, and semantic model is especially big,
Learning training speed is especially slow, low efficiency.
In order to improve the learning training efficiency of semantic model, the present invention proposes a kind of learning training method of semantic model,
The learning training device of semantic model suitable for above-described embodiment.In the present embodiment, compiled by the index to semantic model
Code carries out dimensionality reduction, to improve the learning training speed of semantic model.
Specifically, core word is predefined, establishes the corresponding core word bank of core word, core word includes movie name, drills
Member's name, director names etc., correspondingly core word bank includes movie name library, Yan Yuanku, Dao Yanku, video display verb library etc., wherein
Preset the corresponding unique index coding of dictionary type and the corresponding unique index coding of each individual character of each dictionary.For example,
The default corresponding unique index of individual character " I " is encoded to " 39 ", and the corresponding unique index of individual character " one " is encoded to " 345 ", dictionary class
The corresponding unique index coding " 9 " of type " film ".Based on the corresponding unique index coding of each dictionary type and each individual character
Corresponding unique index coding, establishes semantic model.
Optionally, in order to further enhance the real-time of semantic model, dynamic update is carried out to core word bank.For example, logical
The movie data library for crossing honeybee video carries out dynamic update to core word bank.
Chinese character removes traditional font and has more than common 3000 altogether, in addition more than uncommon 8000, so at most also with regard to more than 8000
A Chinese individual character is also by dimension from several plus core word adds the order of magnitude of individual character to be substantially no more than 9,000 grades after core word
100000 orders of magnitude are reduced to 9,000 orders of magnitude, and the size of semantic model is reduced to 5M or so from several G easily.And since a dictionary is
Corresponding unique index coding, namely say in performer library that Liu De China, schoolmate, the Zhou Xing core words such as speed all correspond to the same index
Coding, in this way, newly-increased core word and dictionary do not influence the final order of magnitude, the order of magnitude of semantic model is stablized 9,000
Grade.Since the order of magnitude of semantic model substantially reduces, the learning training speed of semantic model is greatly improved.
When carrying out semantic model learning training, for training text, firstly, preset core word bank is based on, using double
Word segmentation processing is carried out to training text to maximum matching method, obtains the corresponding matching word segmentation result of training text, wherein matching point
Word result includes the individual character and/or core word after splitting training text.
For example, by taking training text " I will see the film whose schoolmate of that so-and-so acts the leading role " as an example, based on preset
Core word bank carries out word segmentation processing to training text using two-way maximum matching method, obtains the corresponding matching point of the training text
Word result are as follows:
I, and sees, and one, it is a, that, it is a, who, who, who, schoolmate acts the leading role, film
Wherein, including individual character " I ", " wanting " " seeing " " one " " a " " that " " a " " who ", " who ", " who ", " ", Yi Jihe
Heart word " schoolmate ", " protagonist ", " film ".
If matching in word segmentation result includes individual character, for individual character, it is based on semantic model, is determined single in matching word segmentation result
The corresponding unique index coding of word.
If matching in word segmentation result includes core word, for core word, it is first determined the corresponding dictionary type of core word,
For example, matching includes core word " schoolmate ", " protagonist " and " film " in word segmentation result still by taking above-mentioned example as an example, determine
The corresponding dictionary type of core word " schoolmate " is performer library, and the corresponding dictionary type of core word " protagonist " is video display verb library,
The corresponding dictionary type of core word " film " is movie name library.After the corresponding dictionary type of core word has been determined, it is based on language
Adopted model determines the corresponding unique index coding of dictionary type.
Later, it according to the individual character of acquisition and/or the corresponding index coded representation training text of dictionary type, realizes to semanteme
The learning training of model.For example, still by it is above-mentioned enumerate example for, training text is that " I will see that so-and-so whose Zhang Xue
The film that friend acts the leading role ", obtaining matching word segmentation result by word segmentation processing includes individual character " I ", " wanting " " seeing " " one " " a " " that "
" a " " who ", " who ", " who ", " " and core word " schoolmate ", " protagonist ", " film ".Based on semantic model, determine
" I " corresponding unique index coding " 39 ", " will " unique index coding " 8230 " is corresponded to, " seeing " corresponding uniquely index is compiled
Code " 8228 ", " one " corresponding unique index coding " 345 ", " a " corresponding unique index coding " 238 ", " that " is corresponding unique
Index encode " 467 ", " who " corresponding unique index coding " 2490 ", " " corresponding unique index coding " 4611 ", "
The corresponding unique index coding " 2 " of dictionary type belonging to schoolmate ", the corresponding unique index coding of dictionary type belonging to " protagonist "
" 8235 ", the corresponding unique index coding " 9 " of dictionary type belonging to " film ", then by training text " I to see one that who
So-and-so opens the film of schoolmate's protagonist " index of reference coded representation are as follows: [39,8230,8228,345,238,467,238,2490,
2490,2490,2,8235,4611,9].
Scheme provided in this embodiment is based on preset core word bank, using double when carrying out semantic model learning training
Word segmentation processing is carried out to training text to maximum matching method, obtain the corresponding matching word segmentation result of training text (including individual character with/
Or core word), when there are core word, determine the corresponding dictionary type of core word, and according to preset semantic model, obtain single
Word and/or dictionary type corresponding index coding, with according to acquired index coded representation training text, due in semantic mould
Each individual character, dictionary type respectively correspond unique index coding in type, rather than each word corresponds to unique index coding, single
The quantity of word is far smaller than the quantity of word, and the order of magnitude of semantic model substantially reduces, and therefore, greatly improves semantic model
Learning training efficiency.
Further, the learning training method second embodiment that semantic model of the present invention is proposed based on first embodiment,
In the present embodiment, as shown in figure 3, the step S10 includes:
Step S11 carries out word segmentation processing to training text using two-way maximum matching method, it is corresponding to obtain the training text
Individual character and/or word;
Step S12 is based on the core word bank, judges whether institute's predicate is core word when there are word;If so, holding
The row step S20;If it is not, thening follow the steps S13;
Institute's predicate is split as individual character by step S13, obtains the corresponding matching word segmentation result of the training text.
After carrying out word segmentation processing to training text using two-way maximum matching method, the corresponding individual character of training text is obtained
And/or word.If including word, word may be the core word in core word bank, it is also possible to non-core word.In the present embodiment,
When including word in the corresponding matching word segmentation result of training text, it is based on preset core word bank, whether grammatical term for the character is core word.
To the word of then core word, operates as described in the first embodiment, determine the corresponding dictionary type of core word, with
Obtain corresponding index coding.
For not being the word of core word, word is split as individual character, in this way, the corresponding matching participle knot of the training text obtained
Individual character and core word are just only included in fruit.
In a kind of situation, word segmentation processing is carried out to training text using two-way maximum matching method, it is corresponding to obtain training text
Matching word segmentation result in include multiple words, at this point, be based on core word bank, judge whether each word is core word respectively.For
It is not the word of core word in multiple words, which is split as individual character;For being the word of core word in multiple words, core word pair is determined
The dictionary type answered, to obtain corresponding index coding.
Optionally, the step S11 includes:
Step a carries out word segmentation processing to training text using two-way maximum matching method, it is corresponding to obtain the training text
Two-way word segmentation result;
Step b carries out word frequency analysis to the two-way word segmentation result, determines the instruction according to the word frequency data counted in advance
Practice the corresponding individual character of text and/or word.
When carrying out word segmentation processing to training text using two-way maximum matching method, namely Forward Maximum Method is respectively adopted
Method and reverse maximum matching method carry out word segmentation processing to training text, obtain corresponding two-way word segmentation result, including positive participle
As a result with reverse word segmentation result.
For example, carrying out word segmentation processing by taking training text " film about ocean " as an example using Forward Maximum Method method, obtaining
The positive word segmentation result obtained are as follows: about, ocean, film;Using reverse maximum matching method carry out word segmentation processing, acquisition it is inverse
To word segmentation result are as follows: it closes, Yu Haiyang, film.
In the present embodiment, also statistics has the corresponding word frequency data of each word in advance.For example, carrying out word according to search dog corpus
Frequency counts, and obtains corresponding word frequency data.
Forward Maximum Method method and reverse maximum matching method is being respectively adopted, word segmentation processing is carried out to training text, is obtained
After corresponding forward direction word segmentation result and reverse word segmentation result, according to the word frequency data counted in advance, to positive word segmentation result and inverse
Word frequency analysis is carried out to word segmentation result, determines that the corresponding matching word segmentation result of training text is positive word segmentation result, or is reverse
Word segmentation result, namely determine the corresponding individual character of training text and/or word.
Optionally, the step b includes:
Step b1 carries out word frequency to the positive word segmentation result and reverse word segmentation result respectively according to the word frequency data
Analysis obtains the positive corresponding first word frequency total amount of word segmentation result and corresponding second word of the reverse word segmentation result
Frequency total amount;
Step b2, according to the first word frequency total amount and the second word frequency total amount, determine the positive word segmentation result or
The reverse word segmentation result is the matching word segmentation result;
Step b3, the individual character and/or word for including by the matching word segmentation result, is determined as the corresponding list of the training text
Word and/or word.
Optionally, Forward Maximum Method method and reverse maximum matching method is being respectively adopted, training text is being carried out at participle
Reason, after obtaining corresponding positive word segmentation result and reverse word segmentation result, according to the word frequency data counted in advance, respectively to forward direction point
Word result and reverse word segmentation result carry out word frequency analysis, obtain the corresponding word frequency total amount of positive word segmentation result, and reverse
The corresponding word frequency total amount of word segmentation result.For ease of description, the corresponding word frequency total amount of positive word segmentation result is hereafter known as first
The corresponding word frequency total amount of reverse word segmentation result is known as the second word frequency total amount by word frequency total amount.
Later, the first word frequency total amount is compared with the second word frequency total amount, training text pair is determined according to comparison result
The matching word segmentation result answered is positive word segmentation result or reverse word segmentation result, will individual character included by matching word segmentation result and/or
Word is determined as training text and carries out segmenting corresponding individual character and/or word.
For example, still by taking above-mentioned " film about ocean " training text enumerated as an example, obtain positive word segmentation result and
After reverse word segmentation result, according to the word frequency data counted in advance, the word frequency of " about+ocean " is much larger than the word of " pass+Yu Haiyang "
Frequently, accordingly, it is determined that positive word segmentation result is that the corresponding matching word segmentation result of training text namely training text segment are as follows: about,
Ocean, film.
Optionally, the step b1 includes:
Step b11, according to the word frequency data, the individual character and/or the corresponding word of word for including by the positive word segmentation result
Frequency amount adds up, and obtains the corresponding first word frequency total amount of the positive word segmentation result;
Step b12, according to the word frequency data, the individual character for including by the reverse word segmentation result and/or the corresponding word of word
Frequency amount adds up, and obtains the corresponding second word frequency total amount of the reverse word segmentation result.
Optionally, after obtaining corresponding positive word segmentation result and reverse word segmentation result, according to the word frequency number counted in advance
According to, each individual character and/or the corresponding word frequency amount of word that the positive word segmentation result of acquisition includes, and the list for including by positive word segmentation result
Word and/or the corresponding word frequency amount of word add up, and obtain the corresponding first word frequency total amount of positive word segmentation result.And count in advance
Word frequency data, obtain each individual character and/or the corresponding word frequency amount of word that reverse word segmentation result includes, and by reverse word segmentation result packet
The corresponding word frequency amount of the individual character and/or word included adds up, and obtains the corresponding second word frequency total amount of reverse word segmentation result.
Optionally, in a kind of implementation example, the step b2 includes:
When the ratio of the first word frequency total amount and the second word frequency total amount is greater than default ratio, the forward direction is determined
Word segmentation result is the matching word segmentation result;Described in being less than when the ratio of the first word frequency total amount and the second word frequency total amount
When default ratio, determine that the reverse word segmentation result is the matching word segmentation result.
Optionally, the corresponding first word frequency total amount of positive word segmentation result is being got, and inversely word segmentation result is corresponding
After second word frequency total amount, the ratio of the first word frequency total amount and the second word frequency total amount is calculated, the ratio and default ratio are carried out
Compare, if the ratio of the first word frequency total amount and the second word frequency total amount is greater than default ratio, namely illustrates that positive word segmentation result is corresponding
The first word frequency total amount be much larger than the corresponding second word frequency total amount of reverse word segmentation result, at this point, determine forward direction word segmentation result be instruction
Practice the corresponding matching word segmentation result of text.Conversely, if the ratio of the first word frequency total amount and the second word frequency total amount is less than default ratio,
Namely illustrate that the corresponding first word frequency total amount of positive word segmentation result is less than the corresponding second word frequency total amount of reverse word segmentation result, this
When, determine that reverse word segmentation result is the corresponding matching word segmentation result of training text.
Optionally, implement in example at another, the step b2 includes:
When the difference of the first word frequency total amount and the second word frequency total amount is greater than preset difference value, the forward direction is determined
Word segmentation result is the matching word segmentation result;Described in being less than when the difference of the first word frequency total amount and the second word frequency total amount
When preset difference value, determine that the reverse word segmentation result is the matching word segmentation result.
Optionally, implement in example at another, when getting the corresponding first word frequency total amount of positive word segmentation result, with
And after the corresponding second word frequency total amount of reverse word segmentation result, the difference of the first word frequency total amount and the second word frequency total amount is calculated, it will
The difference is compared with preset difference value, if the difference of the first word frequency total amount and the second word frequency total amount is greater than preset difference value, namely
Illustrate that the corresponding first word frequency total amount of positive word segmentation result is much larger than the corresponding second word frequency total amount of reverse word segmentation result, at this point,
Determine that forward direction word segmentation result is the corresponding matching word segmentation result of training text.Conversely, if the first word frequency total amount and the second word frequency are total
The difference of amount is less than preset difference value, at this point, determining that reverse word segmentation result is the corresponding matching word segmentation result of training text.
Scheme provided in this embodiment is carrying out word segmentation processing to training text using two-way maximum matching method, is being instructed
After practicing the corresponding individual character of text and/or word, if wherein there is word, it is based on core word bank, judges whether the word is core word, if
It is not core word, then the word is split as individual character, just only includes individual character and core in the corresponding matching word segmentation result of training text
Therefore word avoids the training study to the non-core word of enormous quantity grade, further improves the learning training of semantic model
Efficiency.
The present invention also provides a kind of computer readable storage medium, the computer-readable recording medium storage has semanteme
Model learning training program, the semantic model learning training program can be executed by one or more than one processor with
In:
Based on preset core word bank, word segmentation processing is carried out to training text using two-way maximum matching method, described in acquisition
The corresponding matching word segmentation result of training text, the matching word segmentation result includes individual character and/or core word;
When there are core word, the corresponding dictionary type of the core word is determined;
According to preset semantic model, the individual character and/or the corresponding index coding of the dictionary type are obtained, with basis
Training text described in the index coded representation, wherein in the semantic model, the corresponding unique index of each individual character is compiled
Code, the corresponding unique index coding of each dictionary type.
Further, following operation is also realized when the semantic model learning training program is executed by processor:
Word segmentation processing is carried out to training text using two-way maximum matching method, obtains the corresponding individual character of the training text
And/or word;
When there are word, it is based on the core word bank, judges whether institute's predicate is core word;
If it is not, institute's predicate is then split as individual character, the corresponding matching word segmentation result of the training text is obtained.
Further, following operation is also realized when the semantic model learning training program is executed by processor:
If the quantity of institute's predicate be it is multiple, be based on the core word bank, judge whether each word is core word respectively;
Word is not core word if it exists, then institute's predicate is split as individual character, obtains the corresponding matching point of the training text
Word result.
Further, following operation is also realized when the semantic model learning training program is executed by processor:
Word segmentation processing is carried out to training text using two-way maximum matching method, it is two-way point corresponding to obtain the training text
Word result;
According to the word frequency data counted in advance, word frequency analysis is carried out to the two-way word segmentation result, determines the training text
Corresponding individual character and/or word.
Further, following operation is also realized when the semantic model learning training program is executed by processor:
According to the word frequency data, word frequency analysis is carried out to the positive word segmentation result and reverse word segmentation result respectively, is obtained
Take the positive corresponding first word frequency total amount of word segmentation result and the corresponding second word frequency total amount of the reverse word segmentation result;
According to the first word frequency total amount and the second word frequency total amount, the positive word segmentation result or described reverse is determined
Word segmentation result is the matching word segmentation result;
By the matching word segmentation result individual character and/or word that include, be determined as the corresponding individual character of the training text and/or
Word.
Further, following operation is also realized when the semantic model learning training program is executed by processor:
When the ratio of the first word frequency total amount and the second word frequency total amount is greater than default ratio, the forward direction is determined
Word segmentation result is the matching word segmentation result;
When the ratio of the first word frequency total amount and the second word frequency total amount is less than the default ratio, described in determination
Reverse word segmentation result is the matching word segmentation result.
Further, following operation is also realized when the semantic model learning training program is executed by processor:
When the difference of the first word frequency total amount and the second word frequency total amount is greater than preset difference value, the forward direction is determined
Word segmentation result is the matching word segmentation result;
When the difference of the first word frequency total amount and the second word frequency total amount is less than the preset difference value, described in determination
Reverse word segmentation result is the matching word segmentation result.
Further, following operation is also realized when the semantic model learning training program is executed by processor:
According to the word frequency data, individual character and/or the corresponding word frequency amount of word that the positive word segmentation result includes are carried out
It is cumulative, obtain the corresponding first word frequency total amount of the positive word segmentation result;
According to the word frequency data, the corresponding word frequency amount of the individual character and/or word for including by the reverse word segmentation result is carried out
It is cumulative, obtain the corresponding second word frequency total amount of the reverse word segmentation result.
Each reality of learning training method of computer readable storage medium specific embodiment of the present invention and above-mentioned semantic model
It is essentially identical to apply example, details are not described herein.
The present embodiment through the above scheme, when carrying out semantic model learning training, is based on preset core word bank, uses
Two-way maximum matching method carries out word segmentation processing to training text, obtains the corresponding matching word segmentation result of training text (including individual character
And/or core word), when there are core word, determine the corresponding dictionary type of core word, and according to preset semantic model, obtain
Individual character and/or dictionary type corresponding index coding are taken, with according to acquired index coded representation training text, due in language
Each individual character, dictionary type respectively correspond unique index coding in adopted model, rather than the corresponding unique index of each word is compiled
Code, the quantity of individual character are far smaller than the quantity of word, and the order of magnitude of semantic model substantially reduces, and therefore, greatly improves semanteme
The learning training efficiency of model.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, method, article or the system that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or system institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, method of element, article or system.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side
Method can be realized by means of software and necessary general hardware platform, naturally it is also possible to by hardware, but in many cases
The former is more preferably embodiment.Based on this understanding, technical solution of the present invention substantially in other words does the prior art
The part contributed out can be embodied in the form of software products, which is stored in one as described above
In storage medium (such as ROM/RAM, magnetic disk, CD), including some instructions are used so that terminal device (it can be mobile phone,
Computer, server or network equipment etc.) execute method described in each embodiment of the present invention.
The above is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (10)
1. a kind of learning training method of semantic model, which is characterized in that the learning training method of the semantic model include with
Lower step:
Based on preset core word bank, word segmentation processing is carried out to training text using two-way maximum matching method, obtains the training
The corresponding matching word segmentation result of text, the matching word segmentation result includes individual character and/or core word;
When there are core word, the corresponding dictionary type of the core word is determined;
According to preset semantic model, the individual character and/or the corresponding index coding of the dictionary type are obtained, according to
Index training text described in coded representation, wherein in the semantic model, the corresponding unique index coding of each individual character, often
The corresponding unique index coding of a dictionary type.
2. the learning training method of semantic model as described in claim 1, which is characterized in that described to be based on preset core word
Library carries out word segmentation processing to training text using two-way maximum matching method, obtains the corresponding matching participle knot of the training text
The step of fruit includes:
Using two-way maximum matching method to training text carry out word segmentation processing, obtain the corresponding individual character of the training text and/or
Word;
When there are word, it is based on the core word bank, judges whether institute's predicate is core word;
If it is not, institute's predicate is then split as individual character, the corresponding matching word segmentation result of the training text is obtained.
3. the learning training method of semantic model as claimed in claim 2, which is characterized in that it is described when there are word, it is based on
The core word bank judges that the step of whether institute's predicate is core word includes:
If the quantity of institute's predicate be it is multiple, be based on the core word bank, judge whether each word is core word respectively;
It is described if it is not, institute's predicate is then split as individual character, the step of obtaining the training text corresponding matching word segmentation result packet
It includes:
Word is not core word if it exists, then institute's predicate is split as individual character, obtains the corresponding matching participle knot of the training text
Fruit.
4. the learning training method of semantic model as claimed in claim 2 or claim 3, which is characterized in that described to use two-way maximum
Matching method carries out word segmentation processing to training text, and the step of obtaining the corresponding individual character of the training text and/or word includes:
Word segmentation processing is carried out to training text using two-way maximum matching method, obtains the corresponding two-way participle knot of the training text
Fruit;
According to the word frequency data counted in advance, word frequency analysis is carried out to the two-way word segmentation result, determines that the training text is corresponding
Individual character and/or word.
5. the learning training method of semantic model as claimed in claim 4, which is characterized in that the two-way word segmentation result includes
Positive word segmentation result and reverse word segmentation result, the word frequency data that the basis counts in advance carry out word to the two-way word segmentation result
Frequency analysis, the step of determining the corresponding individual character of the training text and/or word include:
According to the word frequency data, word frequency analysis is carried out to the positive word segmentation result and reverse word segmentation result respectively, obtains institute
State the corresponding first word frequency total amount of positive word segmentation result and the corresponding second word frequency total amount of the reverse word segmentation result;
According to the first word frequency total amount and the second word frequency total amount, the positive word segmentation result or the reverse participle are determined
It as a result is the matching word segmentation result;
The individual character and/or word for including by the matching word segmentation result, are determined as the corresponding individual character of the training text and/or word.
6. the learning training method of semantic model as claimed in claim 5, which is characterized in that described according to first word frequency
Total amount and the second word frequency total amount, determine the positive word segmentation result or the reverse word segmentation result is matching participle knot
The step of fruit includes:
When the ratio of the first word frequency total amount and the second word frequency total amount is greater than default ratio, the positive participle is determined
It as a result is the matching word segmentation result;
When the ratio of the first word frequency total amount and the second word frequency total amount is less than the default ratio, determine described reverse
Word segmentation result is the matching word segmentation result.
7. the learning training method of semantic model as claimed in claim 5, which is characterized in that described according to first word frequency
Total amount and the second word frequency total amount, determine the positive word segmentation result or reverse word segmentation result is the matching word segmentation result
Step includes:
When the difference of the first word frequency total amount and the second word frequency total amount is greater than preset difference value, the positive participle is determined
It as a result is the matching word segmentation result;
When the difference of the first word frequency total amount and the second word frequency total amount is less than the preset difference value, determine described reverse
Word segmentation result is the matching word segmentation result.
8. the learning training method of semantic model as claimed in claim 5, which is characterized in that described according to the word frequency number
According to respectively to the positive word segmentation result and reverse word segmentation result progress word frequency analysis, the acquisition positive word segmentation result is corresponding
The first word frequency total amount and the reverse word segmentation result corresponding second word frequency total amount the step of include:
According to the word frequency data, individual character and/or the corresponding word frequency amount of word that the positive word segmentation result includes are added up,
Obtain the corresponding first word frequency total amount of the positive word segmentation result;
According to the word frequency data, the corresponding word frequency amount of individual character and/or word that the reverse word segmentation result includes is added up,
Obtain the corresponding second word frequency total amount of the reverse word segmentation result.
9. a kind of learning training device of semantic model, which is characterized in that the learning training device of the semantic model includes: to deposit
Reservoir, processor and the semantic model learning training program that is stored on the memory and can run on the processor;
Such as semanteme of any of claims 1-8 is realized when the semantic model learning training program is executed by the processor
The step of learning training method of model.
10. a kind of computer readable storage medium, which is characterized in that be stored with semantic mould on the computer readable storage medium
Type learning training program is realized when the semantic model learning training program is executed by processor as any in claim 1-8
The step of learning training method of semantic model described in.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810800318.6A CN109033082B (en) | 2018-07-19 | 2018-07-19 | Learning training method and device of semantic model and computer readable storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810800318.6A CN109033082B (en) | 2018-07-19 | 2018-07-19 | Learning training method and device of semantic model and computer readable storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109033082A true CN109033082A (en) | 2018-12-18 |
CN109033082B CN109033082B (en) | 2022-06-10 |
Family
ID=64643694
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810800318.6A Active CN109033082B (en) | 2018-07-19 | 2018-07-19 | Learning training method and device of semantic model and computer readable storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109033082B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112052424A (en) * | 2020-10-12 | 2020-12-08 | 腾讯科技(深圳)有限公司 | Content auditing method and device |
CN113065360A (en) * | 2021-04-16 | 2021-07-02 | 平安国际智慧城市科技股份有限公司 | Word semantic model construction method and device, computer equipment and storage medium |
CN113392651A (en) * | 2020-11-09 | 2021-09-14 | 腾讯科技(深圳)有限公司 | Training word weight model, and method, device, equipment and medium for extracting core words |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104424177A (en) * | 2013-08-26 | 2015-03-18 | 高德软件有限公司 | Method and device for extracting core words |
CN105096942A (en) * | 2014-05-21 | 2015-11-25 | 清华大学 | Semantic analysis method and semantic analysis device |
CN105426539A (en) * | 2015-12-23 | 2016-03-23 | 成都电科心通捷信科技有限公司 | Dictionary-based lucene Chinese word segmentation method |
CN105893353A (en) * | 2016-04-20 | 2016-08-24 | 广东万丈金数信息技术股份有限公司 | Word segmentation method and word segmentation system |
CN106055560A (en) * | 2016-05-18 | 2016-10-26 | 上海申腾信息技术有限公司 | Method for collecting data of word segmentation dictionary based on statistical machine learning method |
CN106407235A (en) * | 2015-08-03 | 2017-02-15 | 北京众荟信息技术有限公司 | A semantic dictionary establishing method based on comment data |
CN106502994A (en) * | 2016-11-29 | 2017-03-15 | 上海智臻智能网络科技股份有限公司 | A kind of method and apparatus of the keyword extraction of text |
CN106776562A (en) * | 2016-12-20 | 2017-05-31 | 上海智臻智能网络科技股份有限公司 | A kind of keyword extracting method and extraction system |
CN106933799A (en) * | 2015-12-31 | 2017-07-07 | 北京四维图新科技股份有限公司 | A kind of Chinese word cutting method and device of point of interest POI titles |
CN106991085A (en) * | 2017-04-01 | 2017-07-28 | 中国工商银行股份有限公司 | The abbreviation generation method and device of a kind of entity |
CN107204184A (en) * | 2017-05-10 | 2017-09-26 | 平安科技(深圳)有限公司 | Audio recognition method and system |
CN108009228A (en) * | 2017-11-27 | 2018-05-08 | 咪咕互动娱乐有限公司 | A kind of method to set up of content tab, device and storage medium |
-
2018
- 2018-07-19 CN CN201810800318.6A patent/CN109033082B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104424177A (en) * | 2013-08-26 | 2015-03-18 | 高德软件有限公司 | Method and device for extracting core words |
CN105096942A (en) * | 2014-05-21 | 2015-11-25 | 清华大学 | Semantic analysis method and semantic analysis device |
CN106407235A (en) * | 2015-08-03 | 2017-02-15 | 北京众荟信息技术有限公司 | A semantic dictionary establishing method based on comment data |
CN105426539A (en) * | 2015-12-23 | 2016-03-23 | 成都电科心通捷信科技有限公司 | Dictionary-based lucene Chinese word segmentation method |
CN106933799A (en) * | 2015-12-31 | 2017-07-07 | 北京四维图新科技股份有限公司 | A kind of Chinese word cutting method and device of point of interest POI titles |
CN105893353A (en) * | 2016-04-20 | 2016-08-24 | 广东万丈金数信息技术股份有限公司 | Word segmentation method and word segmentation system |
CN106055560A (en) * | 2016-05-18 | 2016-10-26 | 上海申腾信息技术有限公司 | Method for collecting data of word segmentation dictionary based on statistical machine learning method |
CN106502994A (en) * | 2016-11-29 | 2017-03-15 | 上海智臻智能网络科技股份有限公司 | A kind of method and apparatus of the keyword extraction of text |
CN106776562A (en) * | 2016-12-20 | 2017-05-31 | 上海智臻智能网络科技股份有限公司 | A kind of keyword extracting method and extraction system |
CN106991085A (en) * | 2017-04-01 | 2017-07-28 | 中国工商银行股份有限公司 | The abbreviation generation method and device of a kind of entity |
CN107204184A (en) * | 2017-05-10 | 2017-09-26 | 平安科技(深圳)有限公司 | Audio recognition method and system |
CN108009228A (en) * | 2017-11-27 | 2018-05-08 | 咪咕互动娱乐有限公司 | A kind of method to set up of content tab, device and storage medium |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112052424A (en) * | 2020-10-12 | 2020-12-08 | 腾讯科技(深圳)有限公司 | Content auditing method and device |
CN112052424B (en) * | 2020-10-12 | 2024-05-28 | 腾讯科技(深圳)有限公司 | Content auditing method and device |
CN113392651A (en) * | 2020-11-09 | 2021-09-14 | 腾讯科技(深圳)有限公司 | Training word weight model, and method, device, equipment and medium for extracting core words |
CN113392651B (en) * | 2020-11-09 | 2024-05-14 | 腾讯科技(深圳)有限公司 | Method, device, equipment and medium for training word weight model and extracting core words |
CN113065360A (en) * | 2021-04-16 | 2021-07-02 | 平安国际智慧城市科技股份有限公司 | Word semantic model construction method and device, computer equipment and storage medium |
CN113065360B (en) * | 2021-04-16 | 2023-02-07 | 平安国际智慧城市科技股份有限公司 | Word semantic model construction method and device, computer equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN109033082B (en) | 2022-06-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109657054B (en) | Abstract generation method, device, server and storage medium | |
CN107204184B (en) | Audio recognition method and system | |
CN106886567B (en) | Microblogging incident detection method and device based on semantic extension | |
CN109902216A (en) | A kind of data collection and analysis method based on social networks | |
CN105512182B (en) | Sound control method and smart television | |
CN109033082A (en) | The learning training method, apparatus and computer readable storage medium of semantic model | |
CN114492831B (en) | Method and device for generating federal learning model | |
CN111209363B (en) | Corpus data processing method, corpus data processing device, server and storage medium | |
CN112631436B (en) | Method and device for filtering sensitive words of input method | |
CN111767393A (en) | Text core content extraction method and device | |
US20230342667A1 (en) | Classification model training method, semantic classification method, device and medium | |
CN114254158B (en) | Video generation method and device, and neural network training method and device | |
CN111858905B (en) | Model training method, information identification device, electronic equipment and storage medium | |
CN106445915A (en) | New word discovery method and device | |
US20230195998A1 (en) | Sample generation method, model training method, trajectory recognition method, device, and medium | |
JP2022116231A (en) | Training method of organism detection model, device, electronic apparatus and storage medium | |
CN106649732A (en) | Information pushing method and device | |
CN113139043B (en) | Question-answer sample generation method and device, electronic equipment and storage medium | |
CN108268443B (en) | Method and device for determining topic point transfer and acquiring reply text | |
CN114218951A (en) | Entity recognition model training method, entity recognition method and device | |
CN114120166A (en) | Video question and answer method and device, electronic equipment and storage medium | |
CN112860995A (en) | Interaction method, device, client, server and storage medium | |
CN110442696B (en) | Query processing method and device | |
CN111382258A (en) | Method and device for determining electronic reading object chapter | |
CN114490969B (en) | Question and answer method and device based on table and electronic equipment |
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 |