CN108491463A - Label determines method and device - Google Patents
Label determines method and device Download PDFInfo
- Publication number
- CN108491463A CN108491463A CN201810180201.2A CN201810180201A CN108491463A CN 108491463 A CN108491463 A CN 108491463A CN 201810180201 A CN201810180201 A CN 201810180201A CN 108491463 A CN108491463 A CN 108491463A
- Authority
- CN
- China
- Prior art keywords
- label
- candidate
- audio
- comment text
- target
- 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.)
- Pending
Links
Abstract
A kind of label of offer of the embodiment of the present invention determines method and device, belongs to computer application technology.This method includes：The comment text of the audio-visual resource of target is obtained, and comment text is segmented, obtains word segmentation result；Participle in word segmentation result is integrated, multiple candidate labels are obtained, the candidate tag set formed to multiple candidate labels screens, and obtains the label of the audio-visual resource of target.The characteristics of determining label to the comment text of audio-visual resource by being then based on user, capable of more reacting audio-visual resource to label obtains the more relevant information with audio-visual resource content convenient for user.In addition, since label can be automatically determined according to comment text, need not manually mark, so as to save manpower, and it is more efficient.
Description
Technical field
The present embodiments relate to computer application technologies, and method and dress are determined more particularly, to a kind of label
It sets.
Background technology
With the gradually development of TV industry, internet television is just with abundanter content and more recreational ornamental
Experience, penetrates into numerous average families.Demand of the user to video class amusing products is also transferred to from initial audio image quality
In higher artificial intelligence and personalized service.User can more pay close attention to the label of video when browsing video, it is desirable to from mark
More video content informations are obtained in label, to determine whether itself will watch the corresponding video of label.It provides in the related technology
Two kinds of labels determine method, and first method is mainly using the actor names in the title of video or video as label.The
Two kinds of methods mainly determine label according to the theme of video by way of manually marking, and video artefacts are labeled as " love
Label as feelings ", " war ".
For above-mentioned first method, due to being using the title of video, the name of performer as label, label can not be portrayed
The characteristics of video content itself, cannot obtain more video content informations for user by label, into without
It selects itself to want the video watched convenient for user.To above-mentioned second method, due to being manually to mark, people is expended to compare
Power.In addition, due to being the characteristics of determining label according to the theme of video, can not equally portray video content itself, for user
For, more video content informations cannot be obtained by label.
Invention content
To solve the above-mentioned problems, the embodiment of the present invention provides one kind and overcoming the above problem or solve at least partly
The label for stating problem determines method and device.
According to a first aspect of the embodiments of the present invention, it provides a kind of label and determines that method, this method include：
The comment text of the audio-visual resource of target is obtained, and comment text is segmented, obtains word segmentation result；
Participle in word segmentation result is integrated, multiple candidate labels are obtained, the time that multiple candidate labels are formed
It selects tag set to be screened, obtains the label of the audio-visual resource of target.
Method provided in an embodiment of the present invention, by obtain the audio-visual resource of target comment text, and to comment text into
Row participle, obtains word segmentation result.Participle in word segmentation result is integrated, multiple candidate labels are obtained, to multiple candidate marks
It signs the candidate tag set formed to be screened, obtains the label of the audio-visual resource of target.By being then based on user to audio-visual money
The comment text in source determines label, so that the characteristics of label can more react audio-visual resource, is convenient for user's acquisition more and shadow
The relevant information of sound resource content.In addition, since label can be automatically determined according to comment text, need not manually mark, to
Manpower can be saved, and more efficient.
The possible realization method of with reference to first aspect the first ties participle in second of possible realization method
Participle in fruit is integrated, and multiple candidate labels are obtained, including：It puts in order according to each participle in word segmentation result, according to
It is secondary that adjacent participle is spliced, obtain multiple candidate labels.
The possible realization method of with reference to first aspect the first, in the third possible realization method, to multiple times
The candidate tag set that label is formed is selected to be screened, including：
For any candidate's label in candidate tag set, obtains and wrapped in the corresponding all comment texts of the audio-visual resource of target
Comment text quantity containing any candidate label, comment text quantity is screened out from candidate tag set and is less than the first predetermined threshold value
Candidate label；And/or
For any candidate's label in candidate tag set, determine that comment text includes any candidate in all audio-visual resources
The audio-visual resource quantity of label deletes candidate mark of the audio-visual resource quantity less than the second predetermined threshold value from candidate tag set
Label；And/or
It is screened out from candidate tag set while not comprising noun, verb and adjectival candidate label；And/or
Each candidate label in candidate tag set is input to probabilistic language model, each candidate label is exported and corresponds to
Language model score value, screened out from candidate tag set language model score value be less than third predetermined threshold value candidate label.
The possible realization method of with reference to first aspect the first, in the 4th kind of possible realization method, to multiple times
The candidate tag set that label is formed is selected to be screened, including：
For any candidate's label in candidate tag set, obtains and wrapped in the corresponding all comment texts of the audio-visual resource of target
Comment text quantity containing any candidate label determines that comment text includes the audio-visual of any candidate label in all audio-visual resources
Resource quantity；
According to comment text quantity, audio-visual resource quantity, including the comment score value of the comment text of any candidate's label and
The part of speech weight of any candidate's label calculates the significance level score value of any candidate label；
According to being ranked up from big to small to the significance level score value of each candidate label, preset quantity is candidate before screening out
Candidate label except label.
The 4th kind of possible realization method with reference to first aspect, in the 5th kind of possible realization method, this method is also
Including：
It is total according to being segmented in target comment text using the comment text comprising any candidate label as target comment text
The position of each keyword and weight in quantity, target comment text, determine the comment score value of target comment text.
The 4th kind of possible realization method with reference to first aspect, in the 6th kind of possible realization method, this method is also
Including：According to the part of speech of each participle and position in any candidate label, the part of speech of each participle in any candidate label is determined
Weight sums the part of speech weight of each participle in any candidate label, obtains the part of speech weight of any candidate label.
According to a second aspect of the embodiments of the present invention, a kind of label determining device is provided, which includes：
Word-dividing mode, the comment text for obtaining the audio-visual resource of target, and comment text is segmented, it is segmented
As a result；
Screening module obtains multiple candidate labels, to multiple candidate marks for being integrated to the participle in word segmentation result
It signs the candidate tag set formed to be screened, obtains the label of the audio-visual resource of target.
According to a third aspect of the embodiments of the present invention, it provides a kind of label and determines equipment, including：
At least one processor；And
At least one processor being connect with processor communication, wherein：
Memory is stored with the program instruction that can be executed by processor, and the instruction of processor caller is able to carry out first party
The label that any possible realization method is provided in the various possible realization methods in face determines method.
According to the fourth aspect of the invention, a kind of non-transient computer readable storage medium, non-transient computer are provided
Readable storage medium storing program for executing stores computer instruction, and computer instruction makes the various possible realization methods of computer execution first aspect
In the label that is provided of any possible realization method determine method.
It should be understood that above general description and following detailed description is exemplary and explanatory, it can not
Limit the embodiment of the present invention.
Description of the drawings
Fig. 1 is that a kind of label of the embodiment of the present invention determines the flow diagram of method；
Fig. 2 is that a kind of label of the embodiment of the present invention determines the flow diagram of method；
Fig. 3 is that a kind of label of the embodiment of the present invention builds the structural schematic diagram of system；
Fig. 4 is a kind of block diagram of label determining device of the embodiment of the present invention；
Fig. 5 is that a kind of label of the embodiment of the present invention determines the block diagram of equipment.
Specific implementation mode
With reference to the accompanying drawings and examples, the specific implementation mode of the embodiment of the present invention is described in further detail.With
Lower embodiment is not limited to the range of the embodiment of the present invention for illustrating the embodiment of the present invention.
It for the audio-visual resource in internet, is browsed for the ease of user, each audio-visual resource can pre-set correspondence
Label.For example, the label of video resource (such as movie resource) can be " war ", " love ", " life ", music sources
Label can be " prevalence ", " rural area ", " absolute music " etc..For said circumstances, an embodiment of the present invention provides a kind of label is true
Method is determined, for the label of the audio-visual resource of determination.It should be noted that the embodiment of the present invention is determined for video resource
Label, can be also used for the label for determining music sources, the embodiment of the present invention is not especially limited this.Wherein, video provides
Source may include movie resource, original video resource etc., and the embodiment of the present invention is also not especially limited this.Referring to Fig. 1, the party
Method includes：101, the comment text of the audio-visual resource of target is obtained, and comment text is segmented, obtains word segmentation result；102、
Participle in word segmentation result is integrated, multiple candidate labels are obtained, the candidate tally set that multiple candidate labels are formed
Conjunction is screened, and the label of the audio-visual resource of target is obtained.
In above-mentioned steps 101, comment text can be evaluation of the different users to the audio-visual resource of target, can by
The mode captured in internet obtains, and is such as captured in question and answer website or audio-visual evaluation website, the embodiment of the present invention to this not
Make specific limit.It is evaluated due to might have multidigit user for the audio-visual resource of target, in internet, to obtain
When taking the comment text of the audio-visual resource of target, a plurality of comment text may be got.Correspondingly, divide to comment text
When word, each comment text can be segmented, obtain the word segmentation result of each comment text.Wherein, comment text can be
The text of Chinese or other languages, the embodiment of the present invention are not especially limited this.
For any comment text, the participle in the word segmentation result of the comment text can be integrated, to be somebody's turn to do
The corresponding multiple candidate labels of comment text.Wherein, Integration Mode can be splicing, or deal with participle (such as text
Originally simplify processing) after merge again, the embodiment of the present invention is not especially limited this.It should be noted that splice every time
It can be 2 to segment quantity, or 3.Correspondingly, 2 participles are may include in the candidate label spliced, it can also
Including 3 participles, the embodiment of the present invention are also not especially limited this.
For example, with the audio-visual resource of target for a film, the wherein comment text for the film is " terrible, thorn
Swash, role deduces " for.The comment text is segmented, it is " terrible ", " stimulation ", " angle that can also obtain word segmentation result
Color ", " deduction ".When splicing participle quantity is 1, then it is as follows can to obtain candidate label difference：" terrible ", " stimulation ",
" role ", " deduction ".When splicing participle quantity is 2, then it is as follows can to obtain candidate label difference：" terrible stimulation ",
" role deduces " etc..
By above-mentioned integration process, for each comment text, it can obtain one and be made of multiple candidate labels
Candidate tag set, by being screened to the candidate label in each candidate tag set, the time that can will be obtained after screening
Select the label that label is final as the audio-visual resource of the target.Wherein, screening mode can be by way of semantic, syntactic structure
It is screened, the embodiment of the present invention is not especially limited this.For example, for " terrible to deduce " in above-mentioned splicing result,
Due to not meeting syntactic structure, and it is semantic incorrect, so as to which candidate's label is screened out.
Method provided in an embodiment of the present invention, by obtain the audio-visual resource of target comment text, and to comment text into
Row participle, obtains word segmentation result.Participle in word segmentation result is integrated, multiple candidate labels are obtained, to multiple candidate marks
It signs the candidate tag set formed to be screened, obtains the label of the audio-visual resource of target.By being then based on user to audio-visual money
The comment text in source determines label, so that the characteristics of label can more react audio-visual resource, is convenient for user's acquisition more and shadow
The relevant information of sound resource content.In addition, since label can be automatically determined according to comment text, need not manually mark, to
Manpower can be saved, and more efficient.
When in view of being segmented to comment text, obtained after being spliced according to sequence of the participle in comment text
Candidate label is better able to grammaticalness structure and semanteme.Content based on above-mentioned principle and above-described embodiment, works as Integration Mode
For splicing when, the embodiment of the present invention additionally provides a kind of participle joining method to obtain candidate label, and this method includes：According to point
Each participle puts in order in word result, successively splices to adjacent participle, obtains multiple candidate labels.
When splicing to the participle in word segmentation result, the participle quantity chosen every time can be N number of, and the present invention is implemented
Example does not make specific limit to splicing selected participle quantity every time.Wherein, the minimum value of N is 1, and maximum occurrences are that participle is tied
The total quantity segmented in fruit.It in order to vivider is described, participle can be chosen from word segmentation result according to sliding window and carried out
Splicing.Wherein, the size of sliding window is the participle quantity that splicing is selected every time.For example, the size when sliding window is
2, and the participle in word segmentation result be followed successively by first participle, second participle, third participle, the 4th segment and the 5th
When participle, then first participle and second participle can be spliced to obtain first candidate label, by second participle with
Third, which segments, is spliced to obtain second candidate label ..., subsequent and so on.
It should be noted that for the word segmentation result of any comment text, it can be according to the size pair of a variety of sliding windows
Participle in the word segmentation result is spliced.If for example, in word segmentation result comprising five participle, and the size of sliding window in addition to
Can also be 3 except 2.Correspondingly, first participle, second participle can be segmented with third and is spliced to obtain the
One candidate label segments second, third participle is spliced to obtain second candidate mark with the 4th participle
Label ... ..., subsequent and so on.As shown in the above, for any comment text, different sliding window sizes is chosen
It can obtain multiple candidate tag sets of the comment text.
Method provided in an embodiment of the present invention, by putting in order according to each participle in word segmentation result, successively to phase
Adjacent participle is spliced, and multiple candidate labels are obtained.Participle is spliced due to that can put in order according to participle, to spell
The candidate label connect can be more in line with syntactic structure and semanteme.
By the content of above-described embodiment it is found that each comment text can correspond to multiple candidate tag sets, and target shadow
The corresponding comment text of sound resource might have it is multiple, in order to enable determine label can reflect the audio-visual money of target as much as possible
The content characteristic in source, to be screened to the candidate label in candidate tag set.Based in above-described embodiment
Hold, the embodiment of the present invention additionally provides a kind of screening technique of candidate label, and this method includes：For appointing in candidate tag set
One candidate label obtains the comment text number for including any candidate label in the corresponding all comment texts of the audio-visual resource of target
Amount screens out the candidate label that comment text quantity is less than the first predetermined threshold value from candidate tag set；And/or for candidate
Any candidate's label in tag set determines the audio-visual number of resources that comment text includes any candidate label in all audio-visual resources
Amount deletes the candidate label that audio-visual resource quantity is less than the second predetermined threshold value from candidate tag set；And/or it is marked from candidate
It is screened out in label set while not comprising noun, verb and adjectival candidate label；And/or it will be every in candidate tag set
One candidate label is input to probabilistic language model, the corresponding language model score value of each candidate label is exported, from candidate tally set
The candidate label that language model score value is less than third predetermined threshold value is screened out in conjunction.Wherein, the first predetermined threshold value, the second predetermined threshold value
And third predetermined threshold value can be configured according to demand, the embodiment of the present invention does not make specific limit to the value of three.
For above-mentioned first way, for any candidate label in candidate tag set, if the audio-visual resource pair of target
There are 100 comment texts in all comment texts answered, and it includes the time to have in this 100 comment texts in 10 comment texts
Label is selected, to which the corresponding comment text quantity of candidate's label is 10.As procedure described above, it may be determined that in candidate tag set
The corresponding comment text quantity of each candidate's label screens out comment text quantity from candidate tag set and is less than the first default threshold
The candidate label of value.
For the above-mentioned second way, for any candidate label in candidate tag set, if there is all audio-visual resources
Include candidate's label in the middle corresponding comment text there are 100 audio-visual resources, to can determine that candidate's label is corresponding
Audio-visual resource quantity is 100.As procedure described above, it may be determined that the corresponding audio-visual resource of each candidate's label in candidate tag set
Quantity deletes the candidate label that audio-visual resource quantity is less than the second predetermined threshold value from candidate tag set.
For the third above-mentioned mode, since the noun in candidate label usually can objectively react the audio-visual resource of target
Particular content, adjective can objectively react impression of the user after the audio-visual resource of browsing objective, and verb can be objectively
React the audio-visual resource of target theme, so as to using noun, adjective and the verb in candidate label as screen when the considerations of
Factor.For example, by taking the audio-visual resource of target is film as an example, noun can be film correlation name, the historical events etc..Adjective
Can be shock, moved, terrible etc., to react the impression after user has watched film.Verb can be to save, revenge etc.,
To react the theme of the film.
As shown in the above description, if not including noun, verb and adjective simultaneously in candidate label, illustrate candidate label
It is probably unrelated with the content of the audio-visual resource of target, such as comment of pouring water, advertising commentary, so as to from candidate tag set
It screens out while not comprising noun, verb and adjectival candidate label.
For above-mentioned 4th kind of mode, probabilistic language model can by advance training obtain, the embodiment of the present invention to this not
Make specific limit.Specifically, trained label can be predefined, and determines the corresponding language model score value of each trained label.It will
Input of the training label as initial model, using the corresponding language model score value of each trained label as the defeated of initial model
Go out, to be trained initial model to obtain probabilistic language model.Based on probabilistic language model, candidate mark can be calculated
The corresponding language model score value of each candidate label, small to screen out language model score value from candidate tag set in label set
In the candidate label of third predetermined threshold value.
It should be noted that above-mentioned four kinds of screening modes are mainly preliminary screening, above-mentioned four kinds of sieves during actual implementation
It selects mode may be performed simultaneously, one of which or several ways can also be selected to execute, the embodiment of the present invention is not made this specifically
It limits.In addition, the execution sequential of above-mentioned four kinds of modes can be configured according to demand during actual implementation, the present invention is implemented
Example is also not especially limited this.In addition, since the audio-visual resource of target may correspond to a plurality of comment text, and every comment is literary
Originally multiple candidate tag sets may be corresponded to can be as procedure described above to every comment text when to practical preliminary screening
Each of candidate tag set screened.
In the above-described embodiments, preliminary screening can be carried out to candidate tag set.In conjunction with the position segmented in candidate label
The part of speech set and segmented can also do further screening to candidate tag set.Correspondingly, the content based on above-described embodiment,
As a kind of alternative embodiment, the embodiment of the present invention additionally provides a kind of screening technique of candidate label.It should be noted that should
The method further screened can execute after above-mentioned preliminary screening, can also be individually performed, and can be combined with above-mentioned several
The mode of kind preliminary screening executes, and the embodiment of the present invention is not especially limited this.Referring to Fig. 2, the method further screened
Including：201, it for any candidate's label in candidate tag set, obtains in the corresponding all comment texts of the audio-visual resource of target
Include the comment text quantity of any candidate's label, determines the shadow that comment text includes any candidate label in all audio-visual resources
Sound resource quantity；202, according to comment text quantity, audio-visual resource quantity, include the comment of the comment text of any candidate's label
The part of speech weight of score value and any candidate label calculates the significance level score value of any candidate label；203, according to from big to small
The significance level score value of each candidate label is ranked up, the candidate label before screening out except the candidate label of preset quantity.
In above-mentioned steps 201, for any candidate label, the corresponding comment text quantity of candidate's label can be first obtained
And audio-visual resource quantity.Wherein, comment text quantity can be indicated with TF, and audio-visual resource quantity can be that DF is indicated.If for example,
Candidate's label occurred in 5 comments of a film (the audio-visual resource of target), then TF values are 5.If there is commenting for 10 films
Occurs candidate's label in paper sheet, then DF values are 10.
The calculating process of above-mentioned steps 202 can be calculated by following formula, and particular content is as follows：
In above-mentioned formula, score1 indicates the significance level score value of candidate label, and TF indicates that candidate label is corresponding comments
By amount of text, DF indicates the corresponding audio-visual resource quantity of candidate label.N indicates the total quantity of all audio-visual resources, wsIt indicates
Include the comment score value of the comment text of candidate's label, wpIndicate the part of speech weight of candidate's label.Wherein, comment score value can
For indicating the degree of correlation between comment text and the audio-visual resource of target comprising candidate's label, comment score value is bigger, then
Show that the degree of correlation between comment text and the audio-visual resource of target comprising candidate's label is higher.Part of speech weight can be used for table
Show description degree of the candidate label to the audio-visual resource of target, part of speech weight is bigger, then illustrates candidate's label to the audio-visual money of target
Have when source is described higher descriptive.It should be noted that as shown in the above, it, can for any candidate label
It includes candidate's label that can have multiple comment texts；At this point, wsCan be the highest comment score value in multiple comment texts,
Or the average value of the comment score value of multiple comment texts, the embodiment of the present invention are not especially limited this.
Now the calculation formula of the significance level score value of above-mentioned candidate label is explained, if candidate label is audio-visual in target
Occur in multiple comment texts of resource, then illustrates that user can more mention candidate mark in the audio-visual resource of comment target
Label further illustrate the characteristics of candidate's label can react target audio-visual resource.Therefore, it in above-mentioned calculation formula, will wait
The corresponding comment text quantity of label is selected then to be waited as multiplier to which the corresponding comment text quantity of candidate label is bigger
Select the significance level score value of label also can be bigger therewith.If candidate label occurs in the comment text of less audio-visual resource,
That is the value of the corresponding audio-visual resource quantity of candidate's label is smaller, then illustrates that candidate's label has enough separating capacities to distinguish
Other audio-visual resources.Therefore, in above-mentioned calculation formula, the value of N/DF can bigger.It is using N/DF as multiplier, then candidate to mark
The significance level score value of label also can be bigger therewith.
If candidate label appears in core comment text (commenting on the larger comment text of score value), illustrate candidate mark
The significance level of label is higher.Correspondingly, using the comment score value of the comment text comprising candidate label as multiplier, when comprising
The comment score value w of the comment text of candidate's labelsWhen bigger, then the significance level score value of candidate label also can be bigger therewith.
For the part of speech weight w of candidate labelp, the weight of noun, adjective and verb occurs with it in candidate label
Position is related, and more rearward, then the part of speech weight of the noun is then higher in candidate label for the position that noun occurs in candidate label.
The position that adjective and verb occur in candidate label is more forward, then in candidate label the adjective and noun part of speech weight
Also higher.Wherein, when determining the part of speech weight of noun, adjective, verb, due to the importance highest of noun, to part of speech
The value of weight can be according to maximum standard setting, and adjective is identical as the importance of verb but important ratio noun is low, to
The value of part of speech weight can be configured according to less than the standard of noun and according to identical standard, and the embodiment of the present invention does not make this
It is specific to limit.
For any candidate label, using the comment text comprising candidate's label as target comment text.Based on above-mentioned
The content of embodiment, as a kind of alternative embodiment, the embodiment of the present invention additionally provides a kind of commenting for determining target comment text
By the method for score value, this method includes：According to each keyword in participle total quantity, target comment text in target comment text
Position and weight, determine the comment score value of target comment text.
Wherein, keyword is pre-set video display label, such as the performer of video display, director, type, country, language.It will
Pre-set keyword is matched with target comment text, it may be determined that the keyword in target comment text.Above-mentioned determination
The process of comment score value can refer to following formula：
In above-mentioned formula, score2 is the comment score value of target comment text, and w is keyword in target comment text
Weight, L indicate that the participle total quantity in target comment text, p are the position of keyword in target comment text.Wherein, crucial
The weight of word can be pre-set, and such as weight of director can be higher than performer, and the weight of protagonist is higher than other performers, the embodiment of the present invention
Specific limit is not made to the set-up mode of weight.As shown from the above formula, keyword position more rearward, then target comment text
It is also bigger to comment on score value.When in target comment text including multiple keywords, each pass can be calculated by above-mentioned formula
The corresponding comment score value of each keyword is summed, can be commented summed result as target by the corresponding comment score value of keyword
The comment score value of paper sheet.It is sorted from big to small to the comment score value of each target comment text, preset quantity before choosing
A target comment text can be used as core comment text.
Content based on above-described embodiment, as a kind of alternative embodiment, the embodiment of the present invention additionally provides a kind of determination
The method of the part of speech weight of candidate label, including：According to the part of speech of each participle and position in any candidate label, determine any
The part of speech weight of each participle in any candidate label is summed, is obtained by the part of speech weight of each participle in candidate label
The part of speech weight of any candidate's label.
By taking audio-visual resource is film as an example, the various functions of above-described embodiment can build system by label as shown in Figure 3
It realizes.Wherein, data Layer can provide basic data for system, the comment text of each film be obtained from network, and store and arrive
In film comment library.Can also include movie contents library in addition, for the ease of searching the relevant information of movie contents.Algorithm layer
Frequency statistics algorithm including basis and text analyzing algorithm provide algorithm support, text for the film label extraction of application layer
Parser includes participle, language model, part of speech analysis etc., and frequency statistics algorithm provides the system of the texts statistical indicators such as TF, DF
Meter method.Application layer passes sequentially through label generation, label primary dcreening operation and label marking, obtains the final label of film.The mark of generation
Label can be stored, and can be used in the different business such as shadow list excavates, film is recommended.By taking audio-visual resource is film as an example, according to
Label determined by method provided in an embodiment of the present invention can be as shown in table 1 below：
Table 1
| Film | Label |
| Mozart passes | Music is very good |
| The mysterious invasion of the Nature | Intensive neurosis |
| The Battle Of Taierzhuang | Best war film |
| The woman of French first lieutenant | It plays in play |
It should be noted that above-mentioned all alternative embodiments, may be used the optional implementation that any combination forms the present invention
Example, this is no longer going to repeat them.
Content based on above-described embodiment, an embodiment of the present invention provides a kind of label determining device, which determines dress
It sets the label for executing in above method embodiment and determines method.Referring to Fig. 4, which includes：
Word-dividing mode 401, the comment text for obtaining the audio-visual resource of target, and comment text is segmented, it obtains
Word segmentation result；
Screening module 402 obtains multiple candidate labels, to multiple times for being integrated to the participle in word segmentation result
It selects the candidate tag set that label is formed to be screened, obtains the label of the audio-visual resource of target.
As a kind of alternative embodiment, screening module 402, for putting in order according to each participle in word segmentation result,
Adjacent participle is spliced successively, obtains multiple candidate labels.
As a kind of alternative embodiment, screening module 402, for for any candidate's label in candidate tag set, obtaining
The comment text quantity for including any candidate label in the corresponding all comment texts of the audio-visual resource of target is taken, from candidate tally set
The candidate label that comment text quantity is less than the first predetermined threshold value is screened out in conjunction；And/or for any time in candidate tag set
Label is selected, the audio-visual resource quantity that comment text includes any candidate label in all audio-visual resources is determined, from candidate tally set
The candidate label that audio-visual resource quantity is less than the second predetermined threshold value is deleted in conjunction；And/or it is screened out simultaneously from candidate tag set
Not comprising noun, verb and adjectival candidate label；And/or each candidate label in candidate tag set is input to
Probabilistic language model exports the corresponding language model score value of each candidate label, language model is screened out from candidate tag set
Score value is less than the candidate label of third predetermined threshold value.
As a kind of alternative embodiment, screening module 402, including：
Determination unit, for for any candidate's label in candidate tag set, obtaining the corresponding institute of the audio-visual resource of target
There is the comment text quantity for including any candidate label in comment text, determines that comment text includes any in all audio-visual resources
The audio-visual resource quantity of candidate label；
Computing unit, for according to comment text quantity, audio-visual resource quantity, including the comment text of any candidate's label
Comment score value and any candidate label part of speech weight, calculate the significance level score value of any candidate label；
Unit is screened out, for according to being ranked up from big to small to the significance level score value of each candidate label, before screening out
Candidate label except preset quantity candidate's label.
As a kind of alternative embodiment, screening module 402 further includes：
Determination unit, for will include any candidate label comment text as target comment text, commented according to target
The position of each keyword and weight in total quantity, target comment text are segmented in paper sheet, determine commenting for target comment text
By score value.
As a kind of alternative embodiment, screening module 402 further includes：
Summation unit, for according to the part of speech of each participle and position in any candidate label, determining any candidate label
In each participle part of speech weight, the part of speech weight of each participle in any candidate label is summed, any candidate is obtained
The part of speech weight of label.
Device provided in an embodiment of the present invention, by obtain the audio-visual resource of target comment text, and to comment text into
Row participle, obtains word segmentation result.Participle in word segmentation result is integrated, multiple candidate labels are obtained, to multiple candidate marks
It signs the candidate tag set formed to be screened, obtains the label of the audio-visual resource of target.By being then based on user to audio-visual money
The comment text in source determines label, so that the characteristics of label can more react audio-visual resource, is convenient for user's acquisition more and shadow
The relevant information of sound resource content.In addition, since label can be automatically determined according to comment text, need not manually mark, to
Manpower can be saved, and more efficient.
An embodiment of the present invention provides a kind of labels to determine equipment.Referring to Fig. 5, which includes：Processor
(processor) 501, memory (memory) 502 and bus 503；
Wherein, processor 501 and memory 502 complete mutual communication by bus 503 respectively；
Processor 501 is used to call the program instruction in memory 502, true to execute the label that above-described embodiment is provided
Determine method, such as including：The comment text of the audio-visual resource of target is obtained, and comment text is segmented, obtains word segmentation result；
Participle in word segmentation result is integrated, multiple candidate labels are obtained, the candidate tally set that multiple candidate labels are formed
Conjunction is screened, and the label of the audio-visual resource of target is obtained.
The embodiment of the present invention provides a kind of non-transient computer readable storage medium, the non-transient computer readable storage medium
Matter stores computer instruction, which makes computer execute the label that above-described embodiment is provided to determine method, such as
Including：The comment text of the audio-visual resource of target is obtained, and comment text is segmented, obtains word segmentation result；To word segmentation result
In participle integrated, obtain multiple candidate labels, the candidate tag set formed to multiple candidate labels screens,
Obtain the label of the audio-visual resource of target.
One of ordinary skill in the art will appreciate that：Realize that all or part of step of above method embodiment can pass through
The relevant hardware of program instruction is completed, and program above-mentioned can be stored in a computer read/write memory medium, the program
When being executed, step including the steps of the foregoing method embodiments is executed；And storage medium above-mentioned includes：ROM, RAM, magnetic disc or light
The various media that can store program code such as disk.
Label described above determines that the embodiments such as equipment are only schematical, wherein illustrate as separating component
Unit may or may not be physically separated, and the component shown as unit may or may not be object
Manage unit, you can be located at a place, or may be distributed over multiple network units.It can select according to the actual needs
Some or all of module therein is selected to achieve the purpose of the solution of this embodiment.Those of ordinary skill in the art are not paying wound
In the case of the labour for the property made, you can to understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It is realized by the mode of software plus required general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on
Stating technical solution, substantially the part that contributes to existing technology can be expressed in the form of software products in other words, should
Computer software product can store in a computer-readable storage medium, such as ROM/RAM, magnetic disc, CD, including several fingers
It enables and using so that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation
Certain Part Methods of example or embodiment.Finally, the present processes are only preferable embodiment, are not intended to limit this
The protection domain of inventive embodiments.It is all the embodiment of the present invention spirit and principle within, made by it is any modification, equally replace
It changes, improve, should be included within the protection domain of the embodiment of the present invention.
Claims (9)
1. a kind of label determines method, which is characterized in that including：
The comment text of the audio-visual resource of target is obtained, and the comment text is segmented, obtains word segmentation result；
Participle in the word segmentation result is integrated, multiple candidate labels are obtained, the multiple candidate label is formed
Candidate tag set screened, obtain the label of the audio-visual resource of the target.
2. according to the method described in claim 1, it is characterized in that, the participle in the word segmentation result is integrated,
Multiple candidate labels are obtained, including：
It puts in order according to each participle in the word segmentation result, adjacent participle is spliced successively, obtains multiple times
Select label.
3. according to the method described in claim 1, it is characterized in that, the candidate mark formed to the multiple candidate label
Label set is screened, including：
For any candidate label in the candidate tag set, the corresponding all comment texts of the audio-visual resource of the target are obtained
In include the comment text quantity of any candidate label, screening out comment text quantity from the candidate tag set is less than
The candidate label of first predetermined threshold value；And/or
For any candidate label in the candidate tag set, determine that comment text includes described any in all audio-visual resources
The audio-visual resource quantity of candidate label deletes audio-visual resource quantity less than the second predetermined threshold value from the candidate tag set
Candidate label；And/or
It is screened out from the candidate tag set while not comprising noun, verb and adjectival candidate label；And/or
Each candidate label in the candidate tag set is input to probabilistic language model, each candidate label is exported and corresponds to
Language model score value, candidate mark of the language model score value less than third predetermined threshold value is screened out from the candidate tag set
Label.
4. according to the method described in claim 1, it is characterized in that, the candidate mark formed to the multiple candidate label
Label set is screened, including：
For any candidate label in the candidate tag set, the corresponding all comment texts of the audio-visual resource of the target are obtained
In include the comment text quantity of any candidate label, determine that comment text includes any time in all audio-visual resources
Select the audio-visual resource quantity of label；
According to the comment text quantity, the audio-visual resource quantity, including the comment text of any candidate label is commented
By the part of speech weight of score value and any candidate label, the significance level score value of any candidate label is calculated；
According to being ranked up from big to small to the significance level score value of each candidate label, the candidate label of preceding preset quantity is screened out
Except candidate label.
5. according to the method described in claim 4, it is characterized in that, further including：
It is total according to being segmented in target comment text using the comment text comprising any candidate label as target comment text
The position of each keyword and weight in quantity, target comment text, determine the comment score value of the target comment text.
6. according to the method described in claim 4, it is characterized in that, further including：
According to the part of speech of each participle and position in any candidate label, each participle in any candidate label is determined
Part of speech weight, the part of speech weight of each participle in any candidate label is summed, any candidate mark is obtained
The part of speech weight of label.
7. a kind of label determining device, which is characterized in that including：
Word-dividing mode, the comment text for obtaining the audio-visual resource of target, and the comment text is segmented, it is segmented
As a result；
Screening module obtains multiple candidate labels, to the multiple time for being integrated to the participle in the word segmentation result
It selects the candidate tag set that label is formed to be screened, obtains the label of the audio-visual resource of the target.
8. a kind of label determines equipment, which is characterized in that including：
At least one processor；And
At least one processor being connect with the processor communication, wherein：
The memory is stored with the program instruction that can be executed by the processor, and the processor calls described program to instruct energy
Enough methods executed as described in claim 1 to 6 is any.
9. a kind of non-transient computer readable storage medium, which is characterized in that the non-transient computer readable storage medium is deposited
Computer instruction is stored up, the computer instruction makes the computer execute the method as described in claim 1 to 6 is any.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810180201.2A CN108491463A (en) | 2018-03-05 | 2018-03-05 | Label determines method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810180201.2A CN108491463A (en) | 2018-03-05 | 2018-03-05 | Label determines method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108491463A true CN108491463A (en) | 2018-09-04 |
Family
ID=63341363
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810180201.2A Pending CN108491463A (en) | 2018-03-05 | 2018-03-05 | Label determines method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108491463A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109840301A (en) * | 2019-01-22 | 2019-06-04 | 珠海天燕科技有限公司 | A kind of method and device |
| CN109840292A (en) * | 2018-12-17 | 2019-06-04 | 北京百度网讯科技有限公司 | The generation method and its device of video tab |
| CN109992646A (en) * | 2019-03-29 | 2019-07-09 | 腾讯科技（深圳）有限公司 | The extracting method and device of text label |
| CN110188356A (en) * | 2019-05-30 | 2019-08-30 | 腾讯音乐娱乐科技（深圳）有限公司 | Information processing method and device |
| CN111079026A (en) * | 2019-11-28 | 2020-04-28 | 精硕科技（北京）股份有限公司 | Method, storage medium and device for determining character impression data |
| CN111881674A (en) * | 2020-06-28 | 2020-11-03 | 百度在线网络技术（北京）有限公司 | Core commodity word mining method and device, electronic equipment and storage medium |
| CN112101012A (en) * | 2020-09-25 | 2020-12-18 | 北京百度网讯科技有限公司 | Interactive domain determining method and device, electronic equipment and storage medium |
| CN112948677A (en) * | 2021-02-26 | 2021-06-11 | 上海携旅信息技术有限公司 | Recommendation reason determination method, system, device and medium based on comment aesthetic feeling |
| CN113595860A (en) * | 2020-04-30 | 2021-11-02 | 阿里巴巴集团控股有限公司 | Data processing method and device, electronic equipment and computer storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101625695A (en) * | 2009-08-20 | 2010-01-13 | 中国科学院计算技术研究所 | Method and system for extracting complex named entities from Web video p ages |
| CN104778209A (en) * | 2015-03-13 | 2015-07-15 | 国家计算机网络与信息安全管理中心 | Opinion mining method for ten-million-scale news comments |
| WO2015196909A1 (en) * | 2014-06-27 | 2015-12-30 | 北京奇虎科技有限公司 | Word segmentation method and device |
| CN106354861A (en) * | 2016-09-06 | 2017-01-25 | 中国传媒大学 | Automatic film label indexing method and automatic indexing system |
| CN107515934A (en) * | 2017-08-29 | 2017-12-26 | 四川长虹电器股份有限公司 | A kind of film semanteme personalized labels optimization method based on big data |
| CN107679217A (en) * | 2017-10-19 | 2018-02-09 | 北京百度网讯科技有限公司 | Association method for extracting content and device based on data mining |
-
2018
- 2018-03-05 CN CN201810180201.2A patent/CN108491463A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101625695A (en) * | 2009-08-20 | 2010-01-13 | 中国科学院计算技术研究所 | Method and system for extracting complex named entities from Web video p ages |
| WO2015196909A1 (en) * | 2014-06-27 | 2015-12-30 | 北京奇虎科技有限公司 | Word segmentation method and device |
| CN104778209A (en) * | 2015-03-13 | 2015-07-15 | 国家计算机网络与信息安全管理中心 | Opinion mining method for ten-million-scale news comments |
| CN106354861A (en) * | 2016-09-06 | 2017-01-25 | 中国传媒大学 | Automatic film label indexing method and automatic indexing system |
| CN107515934A (en) * | 2017-08-29 | 2017-12-26 | 四川长虹电器股份有限公司 | A kind of film semanteme personalized labels optimization method based on big data |
| CN107679217A (en) * | 2017-10-19 | 2018-02-09 | 北京百度网讯科技有限公司 | Association method for extracting content and device based on data mining |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109840292A (en) * | 2018-12-17 | 2019-06-04 | 北京百度网讯科技有限公司 | The generation method and its device of video tab |
| CN109840292B (en) * | 2018-12-17 | 2021-06-08 | 北京百度网讯科技有限公司 | Video tag generation method and device |
| CN109840301A (en) * | 2019-01-22 | 2019-06-04 | 珠海天燕科技有限公司 | A kind of method and device |
| CN109992646B (en) * | 2019-03-29 | 2021-03-26 | 腾讯科技（深圳）有限公司 | Text label extraction method and device |
| CN109992646A (en) * | 2019-03-29 | 2019-07-09 | 腾讯科技（深圳）有限公司 | The extracting method and device of text label |
| CN110188356A (en) * | 2019-05-30 | 2019-08-30 | 腾讯音乐娱乐科技（深圳）有限公司 | Information processing method and device |
| CN110188356B (en) * | 2019-05-30 | 2023-05-19 | 腾讯音乐娱乐科技（深圳）有限公司 | Information processing method and device |
| CN111079026A (en) * | 2019-11-28 | 2020-04-28 | 精硕科技（北京）股份有限公司 | Method, storage medium and device for determining character impression data |
| CN111079026B (en) * | 2019-11-28 | 2023-11-24 | 北京秒针人工智能科技有限公司 | Method, storage medium and device for determining character impression data |
| CN113595860A (en) * | 2020-04-30 | 2021-11-02 | 阿里巴巴集团控股有限公司 | Data processing method and device, electronic equipment and computer storage medium |
| CN113595860B (en) * | 2020-04-30 | 2023-06-13 | 阿里巴巴集团控股有限公司 | Data processing method, device, electronic equipment and computer storage medium |
| CN111881674A (en) * | 2020-06-28 | 2020-11-03 | 百度在线网络技术（北京）有限公司 | Core commodity word mining method and device, electronic equipment and storage medium |
| CN111881674B (en) * | 2020-06-28 | 2023-07-25 | 百度在线网络技术（北京）有限公司 | Core commodity word mining method and device, electronic equipment and storage medium |
| CN112101012A (en) * | 2020-09-25 | 2020-12-18 | 北京百度网讯科技有限公司 | Interactive domain determining method and device, electronic equipment and storage medium |
| CN112101012B (en) * | 2020-09-25 | 2024-04-26 | 北京百度网讯科技有限公司 | Interactive domain determining method and device, electronic equipment and storage medium |
| CN112948677A (en) * | 2021-02-26 | 2021-06-11 | 上海携旅信息技术有限公司 | Recommendation reason determination method, system, device and medium based on comment aesthetic feeling |
| CN112948677B (en) * | 2021-02-26 | 2023-11-03 | 上海携旅信息技术有限公司 | Recommendation reason determining method, system, equipment and medium based on comment aesthetic feeling |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108491463A (en) | Label determines method and device | |
| US11153653B2 (en) | Resource recommendation method, device, apparatus and computer readable storage medium | |
| CN110020437B (en) | Emotion analysis and visualization method combining video and barrage | |
| CN110209843A (en) | Multimedia resource playback method, device, equipment and storage medium | |
| CN108595660A (en) | Label information generation method, device, storage medium and the equipment of multimedia resource | |
| WO2020232796A1 (en) | Multimedia data matching method and device, and storage medium | |
| US20140164371A1 (en) | Extraction of media portions in association with correlated input | |
| CN106776528B (en) | Information processing method and device | |
| CN102216945B (en) | Networking with media fingerprints | |
| CN109558513A (en) | A kind of content recommendation method, device, terminal and storage medium | |
| CN110347866B (en) | Information processing method, information processing device, storage medium and electronic equipment | |
| CN110166802A (en) | Barrage processing method, device and storage medium | |
| US20140161423A1 (en) | Message composition of media portions in association with image content | |
| CN112382295A (en) | Voice recognition method, device, equipment and readable storage medium | |
| US20140163956A1 (en) | Message composition of media portions in association with correlated text | |
| CN113901263B (en) | Label generation method and device for video material | |
| CN105787078B (en) | Multimedia title display method and device | |
| CN113038175B (en) | Video processing method and device, electronic equipment and computer readable storage medium | |
| CN113886568A (en) | Text abstract generation method and device | |
| CN109993570A (en) | A kind of orientation launches the method and system of moving advertising | |
| CN111125384A (en) | Multimedia answer generation method and device, terminal equipment and storage medium | |
| CN117336572A (en) | Video abstract generation method, device, computer equipment and storage medium | |
| CN106028094A (en) | Video content providing method and device and electronic equipment | |
| CN110070385A (en) | Advertising commentary method, apparatus, electronic equipment and storage medium | |
| CN112804580B (en) | Video dotting method 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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180904 |