CN107193962A - A kind of intelligent figure method and device of internet promotion message - Google Patents
A kind of intelligent figure method and device of internet promotion message Download PDFInfo
- Publication number
- CN107193962A CN107193962A CN201710374160.6A CN201710374160A CN107193962A CN 107193962 A CN107193962 A CN 107193962A CN 201710374160 A CN201710374160 A CN 201710374160A CN 107193962 A CN107193962 A CN 107193962A
- Authority
- CN
- China
- Prior art keywords
- text
- general
- picture
- intention
- generic
- 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
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5846—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using extracted text
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/5866—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, manually generated location and time information
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/60—Editing figures and text; Combining figures or text
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Library & Information Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The application provides a kind of intelligent figure method and device of internet promotion message.The embodiment of the present application matches somebody with somebody atlas by obtaining the general of generic text；The generic text is expanded into intention text；According to the correlation with the intention text, concentrate the acquisition general figure of one or more to be combined with the intention text from the general figure, generate internet promotion message.It can avoid in the prior art because the quality of internet promotion message picture is uneven, picture, picture and text description in the presence of a large amount of low values are not inconsistent, the problem of greatly affected user's visual experience, improves the degree of correlation of picture and word, it is ensured that internet promotion message it is original.
Description
【Technical field】
The application is related to the Internet, applications field, more particularly to a kind of intelligence of internet promotion message matches somebody with somebody drawing method and dress
Put.
【Background technology】
At present, with the high speed development of social economy, internet promotion message has become a kind of highly important business
A large amount of internet promotion messages are delivered in the new media platform such as internet by industry publicity and marketing means, enterprise, huge to obtain
Economic well-being of workers and staff.In current internet promotion message field, the promotion message of figure is a kind of shape that user more gladly receives
Formula, and the quality of figure, directly affect the experience of user.
It is uneven with plot quality in existing internet promotion message, there is situations below：
Contain in the picture of a large amount of low values, such as picture public containing contrary to law, social ethics or impairment in watermark, picture
The content of interests, or image content is in the presence of the inconsistency with word description altogether, it is impossible to the embodiment word description of complete and accurate
Content.The visual experience of user is greatly affected, thus needs a kind of intelligence to match somebody with somebody drawing method, the phase of picture and word is improved
Guan Du, it is ensured that internet promotion message it is original.
【The content of the invention】
The many aspects of the application provide a kind of intelligent figure method and device of internet promotion message, to improve mutually
The picture character degree of correlation of networking promotion message.
The one side of the application matches somebody with somebody drawing method there is provided a kind of intelligence of internet promotion message, including：
Obtain the general of generic text and match somebody with somebody atlas；
The generic text is expanded into intention text；
According to the correlation with the intention text, concentrated from the general figure obtain the general figure of one or more with
The intention text is combined, and generates internet promotion message.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the acquisition are led to
With text it is general match somebody with somebody atlas, including：
Search daily record to the whole network user is analyzed and excavated, and the interest main body and/or high frequency for obtaining the whole network user are looked into
Word is ask, is clustered and duplicate removal processing, obtains candidate's text；
Scanned for candidate's text as search keyword in the index of the picture text message pre-established,
Obtain the index with candidate's text matches；
Picture corresponding with the index of candidate's text matches is obtained, using candidate's text as generic text, and
It regard the picture as general figure；
All general figures are combined as general with atlas.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the acquisition are led to
With text it is general match somebody with somebody atlas, including：
Advertising Law is not met wherein by artificial/automatic examination ＆ verification filtering and relates to yellow, gambling, the picture containing watermark.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, it is described will be general
Text expands to intention text, including：
Generic text to be extended is inputted in the generation trigger model that training in advance is obtained, generation is described to be extended to lead to
With the corresponding intention text of text.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the generation are touched
Training in advance is obtained hair model in the following ways：
Training sample is chosen, the training sample includes the positive and negative intention text of generic text and generic text；
Obtain training sample to input into original deep-neural-network, and institute is based on according to the original deep-neural-network
The output result of training sample is stated, along the direction for minimizing loss function, is reversely successively updated in the original deep layer network
The weighting parameters of each layer, to cause loss function minimum.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the basis with
The correlation of the intention text, concentrates the selection general figure of one or more to enter with the intention text from the general figure
Row combination, generates internet promotion message, including：
Extraction of semantics is carried out to the general figure, the picture feature of the general figure is obtained；
Picture and text feature extraction is carried out to generic text and the picture feature of the general figure；
The correlation of the general figure and intention text is obtained according to the picture and text feature and arranged according to correlation
Sequence；
The forward general figure of one or more that will sort is combined with the intention text, and letter is promoted in generation internet
Breath.
The another aspect of the application matches somebody with somebody map device, including acquiring unit, expansion there is provided a kind of intelligence of internet promotion message
Unit and generation unit are opened up, wherein,
The acquiring unit, matches somebody with somebody atlas for obtaining the general of generic text；
The expanding element, for generic text to be expanded into intention text；
The generation unit, for according to the correlation with the intention text, selection one to be concentrated from the general figure
Width or several general figures are combined with the intention text, generate internet promotion message.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the acquisition list
Member includes following subelement：
Candidate's text obtains subelement, is analyzed and is excavated for the search daily record to the whole network user, obtains the whole network use
The interest main body and/or high frequency query word at family, are clustered and duplicate removal processing, obtain candidate's text；
Index obtain subelement, for using candidate's text as search keyword in the picture text envelope pre-established
Scanned in the index of breath, obtain the index with candidate's text matches；
General figure obtains subelement, for obtaining picture corresponding with the index of candidate's text matches, will be described
Candidate's text is used as general figure as generic text, and using the picture；
It is general to combine subelement with atlas, it is general with atlas for all general figures to be combined as.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the acquisition list
Member also includes examination ＆ verification subelement,
Picture for being concentrated to general figure carries out artificial/automatic examination ＆ verification, and filtering does not meet Advertising Law wherein and related to
Yellow, gambling, the picture containing watermark.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the extension list
Member includes intention text and extends subelement, is used for,
Generic text to be extended is inputted in the generation trigger model that training in advance is obtained, generation is described to be extended to lead to
With the corresponding intention text of text.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the extension list
Member includes production trigger model and trains subelement, is used for,
Training sample is chosen, the training sample includes the positive and negative intention text of generic text and generic text；
Obtain training sample to input into original deep-neural-network, and institute is based on according to the original deep-neural-network
The output result of training sample is stated, along the direction for minimizing loss function, is reversely successively updated in the original deep layer network
The weighting parameters of each layer, to cause loss function minimum.
Aspect as described above and any possible implementation, it is further provided a kind of implementation, the generation list
Member includes following subelement：
Extraction of semantics subelement, for carrying out extraction of semantics according to the general figure using picture and text relationship type model；
Picture and text feature extraction subelement, is carried to generic text and the picture feature of general figure progress picture and text feature
Take；
Correlation obtains subelement, related to intention text for obtaining the general figure according to the picture and text feature
Property is simultaneously ranked up according to correlation；
Internet promotion message generation subelement, for by the forward general figure of one or more and the intention of sorting
Text is combined, and generates internet promotion message.
There is provided a kind of equipment for the another aspect of the application, it is characterised in that the equipment includes：
One or more processors；
Storage device, for storing one or more programs,
When one or more of programs are by one or more of computing devices so that one or more of processing
Device realizes any above-mentioned method.
The another aspect of the application is stored thereon with computer program there is provided a kind of computer-readable recording medium, and it is special
Levy and be, the program realizes any above-mentioned method when being executed by processor.
From the technical scheme, the embodiment of the present application matches somebody with somebody atlas by obtaining the general of generic text；Will be described logical
Intention text is expanded to text；According to the correlation with the intention text, a width or many for selection is concentrated from general figure
General figure is combined with the intention text, generates internet promotion message.It can avoid keeping away in the prior art
Exempt from the prior art because the quality of internet promotion message picture is uneven, there is the picture of a large amount of low values, picture and text and retouch
State and be not inconsistent, the problem of greatly affected user's visual experience improves the degree of correlation of picture and word, it is ensured that internet
Promotion message it is original.
【Brief description of the drawings】
, below will be to embodiment or description of the prior art in order to illustrate more clearly of the technical scheme in the embodiment of the present application
In required for the accompanying drawing that uses be briefly described, it should be apparent that, drawings in the following description are some realities of the application
Example is applied, for those of ordinary skill in the art, without having to pay creative labor, can also be attached according to these
Figure obtains other accompanying drawings.
The intelligence for the internet promotion message that Fig. 1 provides for the embodiment of the application one matches somebody with somebody the schematic flow sheet of drawing method；
The intelligence for the internet promotion message that Fig. 2 provides for another embodiment of the application matches somebody with somebody the structural representation of map device；
Fig. 3 is suitable for for the block diagram for the exemplary computer system/server for realizing the embodiment of the present invention.
【Embodiment】
To make the purpose, technical scheme and advantage of the embodiment of the present application clearer, below in conjunction with the embodiment of the present application
In accompanying drawing, the technical scheme in the embodiment of the present application is clearly and completely described, it is clear that described embodiment is
Some embodiments of the present application, rather than whole embodiments.Based on the embodiment in the application, those of ordinary skill in the art
The whole other embodiments obtained under the premise of creative work is not made, belong to the scope of the application protection.
In addition, the terms "and/or", only a kind of incidence relation for describing affiliated partner, represents there may be
Three kinds of relations, for example, A and/or B, can be represented：Individualism A, while there is A and B, these three situations of individualism B.Separately
Outside, character "/" herein, it is a kind of relation of "or" to typically represent forward-backward correlation object.
The intelligence for the internet promotion message that Fig. 1 provides for the embodiment of the application one matches somebody with somebody the schematic flow sheet of drawing method, such as
Shown in Fig. 1, comprise the following steps：
101st, obtain the general of generic text and match somebody with somebody atlas；
102nd, the generic text is expanded into intention text；
103rd, according to and the intention text correlation, concentrate selection from the general figure one or more be general and match somebody with somebody
Figure is combined with the intention text, generates internet promotion message.
Wherein, step 101, specifically, it is described general to match somebody with somebody by the general figure of picture searching technical limit spacing generic text
The collection of figure is collectively referred to as general with atlas, including following sub-step：
201st, the search daily record to the whole network user is analyzed and excavated, and obtains the interest main body and/or height of the whole network user
Frequency query word, is clustered and duplicate removal processing, obtains candidate's text；
202nd, candidate's text is carried out in the index of the picture text message pre-established as search keyword
Search, obtains the index with candidate's text matches；
203rd, picture corresponding with the index of candidate's text matches is obtained, candidate's text is regard as general text
This, and it regard the picture as general figure.
204th, all general figures are combined as general with atlas.
It is preferred that, if not obtaining the index with candidate's text matches, operation terminates.
In a kind of implementation of the present embodiment, further, search engine can be to the figure of the search engine collecting
Piece carries out content analysis, obtains the text message of the picture；The text message of the picture is arranged, the figure is set up
The index of the text message of piece.
In another implementation of the present embodiment, further, search engine can also be to the search engine collecting
Picture carry out content analysis and the text to webpage where the picture is analyzed, the text message of the acquisition picture；
The text message of the picture is arranged, the index of the text message of the picture is set up.
It is preferred that, in the object information in the picture of search engine collecting, landscape information, people information and text information
At least one analyzed, that is, analyzed from the angle of image content, including but not limited to analysis picture in thing
At least one of information such as body information, landscape information, people information and text information.
It is preferred that, the text in the structured field of webpage where the picture is analyzed, net where the picture
Text around the web page title field of the structured field including the webpage of page, picture, Text region is carried out to picture obtain
Text, the anchor texts (link Anchor Text) of picture and/or the picture attribute field arrived.
It is preferred that, search engine carries out the processing such as cutting word, cluster, duplicate removal to the text message of the picture.
It is preferred that, the picture that the general figure is concentrated by artificial/automatic examination ＆ verification filtering do not meet wherein Advertising Law with
And yellow, gambling, the picture containing watermark are related to, to ensure the quality of picture in itself.
It is preferred that, automatic examination ＆ verification is wherein to relate to yellow, gambling, the figure containing watermark by image-recognizing method examination ＆ verification filtering
Piece, to ensure the quality of picture in itself, such as using skin color detector and gesture detector, user or to the colour of skin and posture
Carry out signature analysis, feature extraction and match whether judgement picture is the porny for relating to Huang with the similitude of feature.
The whole network user in the present embodiment can include the user of internet, namely internet user in search engine or
When submitting on-line query request in searcher, the server of search engine or searcher can generate corresponding search day
Will, and the application can collect search daily record by the server of all search engines or searcher from internet, be searched for
Daily record.
The step 102, specifically, chooses the generic text for having set up corresponding relation and intention text of magnanimity as instruction
Practice sample to be trained the original deep-neural-network (DNN, Deep Neural Network) built in advance, that is, input
One end can be generic text, and the other end of input is intention text；Deep neural network is from above-mentioned magnanimity training data middle school
Semantic knowledge is practised, the training to the original neural network model is completed, obtains production trigger model (GTM, Generative
Triger Model).Input generic text to be extended in the generation trigger model that training in advance is obtained, wait to expand described in generation
The corresponding intention text of generic text of exhibition.The intention text covers the content of generic text, and will be retouched by semantic extension
State scope expansion.
Training in advance is obtained the generation trigger model in the following ways：
301st, training sample is chosen；
Taken in view of the substantial amounts of training sample of Direct Mark it is huge, and standard be difficult to unification.In the present embodiment, use
The positive and negative intention text of generic text and generic text is as the training sample of training sample, i.e., one by a generic text and two
The pair that individual intention text is constituted is to composition, and one intention text of the pair centerings is more general relative to this than another intention text
More preferably, the two intention texts are referred to as positive example and negative example to the matching relationship of text.Will the training sample it is specifically excellent
Turn to：By generic text, and positive example intention text corresponding with generic text difference and negative example intention text are constituted
Positive and negative training pair.
Wherein it is possible to by manually determining and leading to according to the matching degree between different intention texts and a generic text
With the corresponding positive example intention text of text and negative example intention text, but it is due to required when training original deep-neural-network
Training sample number it is larger, it is necessary to larger human cost be put into, further, since the evaluation mark of the degree of correlation of different people
It is accurate also different, one of the present embodiment preferred embodiment in, can be automatically determined according to the search click logs of user
Corresponding with a query formulation positive example intention text and negative example intention text, for example, entering when user inputs a generic text
After row search, the web page title that user is clicked on based on search result as positive example intention text corresponding with the generic text,
The web page title that user is not clicked on is used as negative example intention text corresponding with the generic text.
302nd, training sample is obtained to input into the original deep-neural-network, and according to the original deep layer nerve net
Output result of the network based on the training sample, along the direction for minimizing loss function, reversely successively updates the original depth
The weighting parameters of each layer in layer network, to cause loss function minimum.
In the present embodiment, due to by it is described it is positive and negative training to completing the training to the original deep-neural-network,
In order to improve training effectiveness, it can preferably construct two completion original deep-neural-networks of identical and be respectively used to receive by general
The positive training pair that text, positive example intention text are constituted, and the negative training pair being made up of generic text, negative example intention text, enter
And realize quick, real-time model training.
It is preferred that operations described below can be included：
By the generic text and with the positive example intention text input to the original deep-neural-network structure
In identical first network, and obtain the first predicted value of the first network output；
By the generic text and with the negative example intention text input to the first network structure identical the
In two networks, and obtain the second predicted value of the second network output；
According to first predicted value, second predicted value and the positive example intention text and the negative example intention text
Correlation partial order between this, counting loss function；
Setting right value update algorithm is taken, along the direction for minimizing loss function, first net is reversely successively updated
The weighting parameters of each layer in network and second network.
Such method is referred to as BP (Back Propagation, backpropagation) algorithm, and specific right value update algorithm has
Various gradient descent methods, such as LBFGS (Quasi-Newton algorithm), or SGD (stochastic gradient descent) etc., wherein SGD convergence rates
Faster, using more.
The difference trained with general deep-neural-network is：The technical scheme of the present embodiment has two identical networks,
Parameter is shared, and the renewal of weights is synchronous all the time.
Judge whether to reach training termination condition set in advance：If：The original deep layer nerve completed will be trained
Network is used as the generation trigger model；Otherwise, return and proceed adjustment.
In the present embodiment, training termination condition can be set according to the actual requirements, for example, training rounds (for example,
1000 times, or 2000 is inferior) or neutral net to aggregated error value of training sample etc., the present embodiment is to this and without limit
System.
After the training of generation trigger model is completed, in addition to step 303, the generation trigger mode that is obtained in training in advance
Generic text to be extended, the generation corresponding intention text of generic text to be extended are inputted in type.
For example, inputting generic text " kitchen finishing " in the generation trigger model after completing training；Generate trigger model
The generic text " kitchen finishing " of input is generated as intention text, such as " open kitchen finishing design sketch ", " 2017 kitchens are filled
Repair design quotation complete works of ", " whole kitchen which good ", " kitchen finishing design sketch specialty customization cabinet ", " modern kitchen finishing effect
Fruit is schemed " etc..
The step 103 specifically includes following sub-step,
401st, extraction of semantics is carried out according to the general figure, obtains the picture feature of the general figure；
First, low-level image feature extraction of semantics is carried out to the general figure, and formed after high dimensional feature vector space, carried out
Image is split, and obtains the element of the general figure.
Specifically, the feature includes color characteristic (such as histogram, accumulative histogram or local histogram), texture
Feature (such as statistic law, Spectrum Method or Structure Method), shape facility (such as area, girth or flex point number), and it is other effective
Feature.In this, in order to prevent some feature weights excessive, it is necessary to which features described above is normalized.Meanwhile, it is special to these
Progress reasonably fractionation, combined treatment are levied, high dimensional feature vector space is formed.In the present embodiment, specific feature extracting method,
The fractionation and combination of the normalized, feature of feature weight can use any existing mature technology, but the present invention is right
This is not limited.
Next, utilizing the high dimensional feature vector extracted, the classification that has supervision of the training based on SVMs (SVM)
Device, obtains the optimum segmentation curved surface of element, image is split.Simultaneously, it is necessary to the background area of acquisition more especially
Discrete fritter, the method merged using region is spliced as a complete target area.In this, grader also may be used
With using other machine learning models, including region Fusion etc., but the present invention is not limited to this.
Finally, will be abstract to semantic concept layer after the element progress elemental recognition of the general figure, obtain described general
The picture feature of figure.
Specifically, each independent territory element as element obtained after image segmentation.Element is that have specific thing
The individual of meaning is managed, by elemental recognition, the implication of element is illustrated with specific word.Specific recognition methods is to pass through
Query image-word mapping table, key values are element image, using the image search method accurately matched.Wherein, image-
Word mapping table is built by network automatic mining, with it, image can be transformed into explanatory note.Connect down
Come, it is according to the comment of acquisition that its is abstract to semantic concept layer.
402nd, picture and text feature extraction is carried out to generic text and the picture feature of the general figure.
403rd, the picture and text correlation model obtained using training in advance, according to the picture and text feature obtain the general figure with
The correlation of intention text, and the general figure is ranked up according to correlation；
Specifically, training in advance is obtained the picture and text correlation model in the following ways：
An original deep-neural-network is built first, and the input of the original deep-neural-network is intention text and logical
With the picture and text feature of figure, the relativity measurement value between the intention text and the picture feature of general figure is output as, is selected
Take training sample to be trained the original deep-neural-network, may finally train and obtain picture and text correlation models；
The training sample is intention text and related picture feature；Training sample is obtained to input to original deep layer nerve
In network, and according to the output result of the original deep-neural-network based on the training sample, letter is lost along minimizing
Several directions, reversely successively updates the weighting parameters of each layer in the original deep layer network, to cause loss function minimum.
It is related to the picture and text by the way that the intention extended in step 102 text and respectively general figure to be sorted are inputted
Property model, relativity measurement value that can respectively between general figure to be sorted and the intention text, and then based on institute
Relativity measurement value is stated, general figure respectively to be sorted is ranked up.
It is preferred that, the parameter of picture and text correlation models is adjusted according to word frequency characteristic and contextual information.
404th, the forward general figure of one or more that will sort is combined with the intention text, and generation internet is pushed away
Guangxin ceases.
The technical scheme provided using the present embodiment, can be avoided in the prior art due to internet promotion message picture
Quality is uneven, there is the picture of a large amount of low values, picture and text description and is not inconsistent, greatly affected the visual experience of user
Problem, ensure that the matching degree of general figure and intention text, and general match somebody with somebody to as much as possible for intention textual association
Figure.
It should be noted that for foregoing each method embodiment, in order to be briefly described, therefore it is all expressed as a series of
Combination of actions, but those skilled in the art should know, the application is not limited by described sequence of movement because
According to the application, some steps can be carried out sequentially or simultaneously using other.Secondly, those skilled in the art should also know
Know, embodiment described in this description belongs to preferred embodiment, involved action and module not necessarily the application
It is necessary.
In the described embodiment, the description to each embodiment all emphasizes particularly on different fields, and does not have the portion being described in detail in some embodiment
Point, it may refer to the associated description of other embodiment.
The intelligence for the internet promotion message that Fig. 2 provides for another embodiment of the application matches somebody with somebody the structural representation of map device,
As indicated with 2, including acquiring unit 21, expanding element 22 and generation unit 23：
Acquiring unit, matches somebody with somebody atlas for obtaining the general of generic text；
Expanding element, for generic text to be expanded into intention text；
Generation unit, for according to and the intention text correlation, from the general figure concentrate one width of selection or
Several general figures are combined with the intention text, generate internet promotion message.
The acquiring unit, it is specifically, described logical for the general figure by picture searching technical limit spacing generic text
It is collectively referred to as with the collection of figure general with atlas.
The acquiring unit includes following subelement,
Candidate's text obtains subelement, is analyzed and is excavated for the search daily record to the whole network user, obtains the whole network use
The interest main body and/or high frequency query word at family, are clustered and duplicate removal processing, obtain candidate's text；
Index obtain subelement, for using candidate's text as search keyword in the picture text envelope pre-established
Scanned in the index of breath, obtain the index with candidate's text matches；
General figure obtains subelement, for obtaining picture corresponding with the index of candidate's text matches, will be described
Candidate's text is used as general figure as generic text, and using the picture；
It is general to combine subelement with atlas, it is general with atlas for all general figures to be combined as.
It is preferred that, if not obtaining the index with candidate's text matches, operation terminates.
In a kind of implementation of the present embodiment, further, search engine can be to the figure of the search engine collecting
Piece carries out content analysis, obtains the text message of the picture；The text message of the picture is arranged, the figure is set up
The index of the text message of piece.
In another implementation of the present embodiment, further, search engine can also be to the search engine collecting
Picture carry out content analysis and the text to webpage where the picture is analyzed, the text message of the acquisition picture；
The text message of the picture is arranged, the index of the text message of the picture is set up.
It is preferred that, in the object information in the picture of search engine collecting, landscape information, people information and text information
At least one analyzed.Namely analyzed from the angle of image content, including but not limited to analyze the thing in picture
At least one of information such as body information, landscape information, people information and text information.
It is preferred that, the text in the structured field of webpage where the picture is analyzed, net where the picture
Text around the web page title field of the structured field including the webpage of page, picture, Text region is carried out to picture obtain
Text, the anchor texts (link Anchor Text) of picture and/or the picture attribute field arrived.
Preferential, search engine carries out the processing such as cutting word, cluster, duplicate removal to the text message of the picture.
It is preferred that, the acquiring unit also includes examination ＆ verification subelement, for the picture concentrated to general figure carry out manually/
Automatic examination ＆ verification, filtering does not meet Advertising Law wherein and relates to yellow, gambling, the picture containing watermark, to ensure the matter of picture in itself
Amount.
It is preferred that, automatic examination ＆ verification is wherein to relate to yellow, gambling, the figure containing watermark by image-recognizing method examination ＆ verification filtering
Piece, to ensure the quality of picture in itself, such as using skin color detector and gesture detector, user or to the colour of skin and posture
Carry out signature analysis, feature extraction and match whether judgement picture is the porny for relating to Huang with the similitude of feature.
The whole network user in the present embodiment can include the user of internet, namely internet user in search engine or
When submitting on-line query request in searcher, the server of search engine or searcher can generate corresponding search day
Will, and the application can collect search daily record by the server of all search engines or searcher from internet, be searched for
Daily record.
The expanding element, specifically, the generic text for having set up corresponding relation and intention text for choosing magnanimity
The original deep-neural-network (DNN, Deep Neural Network) built in advance is trained as training sample, i.e.,
One end of input can be generic text, and the other end of input is intention text；Deep neural network trains number from above-mentioned magnanimity
According to learning semantic knowledge, complete the training to the original neural network model, obtain production trigger model (GTM,
Generative Triger Model).Generic text to be extended is inputted in the generation trigger model that training in advance is obtained,
The generation corresponding intention text of generic text to be extended.The intention text covers the content of generic text, and passes through
Semantic extension will describe scope and expand.
The expanding element includes production trigger model training subelement, intention text extension subelement；
The production trigger model training subelement is used for, and chooses training sample；
Taken in view of the substantial amounts of training sample of Direct Mark it is huge, and standard be difficult to unification.In the present embodiment, use
The positive and negative intention text of generic text and generic text is as the training sample of training sample, i.e., one by a generic text and two
The pair that individual intention text is constituted is to composition, and one intention text of the pair centerings is more general relative to this than another intention text
More preferably, the two intention texts are referred to as positive example and negative example to the matching relationship of text.Will the training sample it is specifically excellent
Turn to：By generic text, and positive example intention text corresponding with generic text difference and negative example intention text are constituted
Positive and negative training pair.
Wherein it is possible to by manually determining and leading to according to the matching degree between different intention texts and a generic text
With the corresponding positive example intention text of text and negative example intention text, but it is due to required when training original deep-neural-network
Training sample number it is larger, it is necessary to larger human cost be put into, further, since the evaluation mark of the degree of correlation of different people
It is accurate also different, one of the present embodiment preferred embodiment in, can be automatically determined according to the search click logs of user
Corresponding with a query formulation positive example intention text and negative example intention text, for example, entering when user inputs a generic text
After row search, the web page title that user is clicked on based on search result as positive example intention text corresponding with the generic text,
The web page title that user is not clicked on is used as negative example intention text corresponding with the generic text.
The production trigger model training subelement is additionally operable to
Training sample is obtained to input into the original deep-neural-network, and according to the original deep-neural-network base
In the output result of the training sample, along the direction for minimizing loss function, the original deep layer net is reversely successively updated
The weighting parameters of each layer in network, to cause loss function minimum.
In the present embodiment, due to by it is described it is positive and negative training to completing the training to the original deep-neural-network,
In order to improve training effectiveness, it can preferably construct two completion original deep-neural-networks of identical and be respectively used to receive by general
The positive training pair that text, positive example intention text are constituted, and the negative training pair being made up of generic text, negative example intention text, enter
And realize quick, real-time model training.
It is preferred that operations described below can be included：
By the generic text and with the positive example intention text input to the original deep-neural-network structure
In identical first network, and obtain the first predicted value of the first network output；
By the generic text and with the negative example intention text input to the first network structure identical the
In two networks, and obtain the second predicted value of the second network output；
According to first predicted value, second predicted value and the positive example intention text and the negative example intention text
Correlation partial order between this, counting loss function；
Setting right value update algorithm is taken, along the direction for minimizing loss function, first net is reversely successively updated
The weighting parameters of each layer in network and second network.
Such method is referred to as BP (Back Propagation, backpropagation) algorithm, and specific right value update algorithm has
Various gradient descent methods, such as LBFGS (Quasi-Newton algorithm), or SGD (stochastic gradient descent) etc., wherein SGD convergence rates
Faster, using more.
The difference trained with general deep-neural-network is：The technical scheme of the present embodiment has two identical networks,
Parameter is shared, and the renewal of weights is synchronous all the time.
Judge whether to reach training termination condition set in advance：If：The original deep layer nerve completed will be trained
Network is used as the generation trigger model；Otherwise, return and proceed adjustment.
In the present embodiment, training termination condition can be set according to the actual requirements, for example, training rounds (for example,
1000 times, or 2000 is inferior) or neutral net to aggregated error value of training sample etc., the present embodiment is to this and without limit
System.
Intention text extends subelement, and to be extended for being inputted in the generation trigger model that training in advance is obtained is general
Text, the generation corresponding intention text of generic text to be extended.
For example, inputting generic text " kitchen finishing " in the generation trigger model after completing training；Generate trigger model
The generic text " kitchen finishing " of input is generated as intention text, such as " open kitchen finishing design sketch ", " 2017 kitchens are filled
Repair design quotation complete works of ", " whole kitchen which good ", " kitchen finishing design sketch specialty customization cabinet ", " modern kitchen finishing effect
Fruit is schemed " etc..
The generation unit, specifically, including following subelement：
Extraction of semantics subelement, for carrying out extraction of semantics according to the general figure, obtains the figure of the general figure
Piece feature；
First, low-level image feature extraction of semantics is carried out to the general figure, and formed after high dimensional feature vector space, carried out
Image is split, and obtains the element of the general figure.
Specifically, the feature includes color characteristic (such as histogram, accumulative histogram or local histogram), texture
Feature (such as statistic law, Spectrum Method or Structure Method), shape facility (such as area, girth or flex point number), and it is other effective
Feature.In this, in order to prevent some feature weights excessive, it is necessary to which features described above is normalized.Meanwhile, it is special to these
Progress reasonably fractionation, combined treatment are levied, high dimensional feature vector space is formed.In the present embodiment, specific feature extracting method,
The fractionation and combination of the normalized, feature of feature weight can use any existing mature technology, but the present invention is right
This is not limited.
Next, utilizing the high dimensional feature vector extracted, the classification that has supervision of the training based on SVMs (SVM)
Device, obtains the optimum segmentation curved surface of element, image is split.Simultaneously, it is necessary to the background area of acquisition more especially
Discrete fritter, the method merged using region is spliced as a complete target area.In this, grader also may be used
With using other machine learning models, including region Fusion etc., but the present invention is not limited to this.
Finally, will be abstract to semantic concept layer after the element progress elemental recognition of the general figure, obtain described general
The picture feature of figure.
Specifically, each independent territory element as element obtained after image segmentation.Element is that have specific thing
The individual of meaning is managed, by elemental recognition, the implication of element is illustrated with specific word.Specific recognition methods is to pass through
Query image-word mapping table, key values are element image, using the image search method accurately matched.Wherein, image-
Word mapping table is built by network automatic mining, with it, image can be transformed into explanatory note.Connect down
Come, it is according to the comment of acquisition that its is abstract to semantic concept layer.
Picture and text feature extraction subelement, is carried to generic text and the picture feature of general figure progress picture and text feature
Take.
Correlation obtains subelement, for the picture and text correlation model obtained using training in advance, according to the picture and text feature
The correlation of the general figure and intention text is obtained, and the general figure is ranked up according to correlation.
Specifically, training in advance is obtained the picture and text correlation model in the following ways：
An original deep-neural-network is built first, and the input of the original deep-neural-network is intention text and logical
With the picture and text feature of figure, the relativity measurement value between the intention text and the picture feature of general figure is output as, is selected
Take training sample to be trained the original deep-neural-network, may finally train and obtain picture and text correlation models；
The training sample is intention text and related picture feature；Training sample is obtained to input to original deep layer nerve
In network, and according to the output result of the original deep-neural-network based on the training sample, letter is lost along minimizing
Several directions, reversely successively updates the weighting parameters of each layer in the original deep layer network, to cause loss function minimum.
Inputted by the intention text that extends with respectively general figure to be sorted to the picture and text correlation models, can be with
The respectively relativity measurement value between general figure to be sorted and the intention text, and then based on the relativity measurement
Value, general figure respectively to be sorted is ranked up.
It is preferred that, the parameter of picture and text correlation models is adjusted according to word frequency characteristic and contextual information.
Internet promotion message generation subelement, for by the forward general figure of one or more and the intention of sorting
Text is combined, and generates internet promotion message.
The technical scheme provided using the present embodiment, can be avoided in the prior art due to internet promotion message picture
Quality is uneven, there is the picture of a large amount of low values, picture and text description and is not inconsistent, greatly affected the visual experience of user
Problem, ensure that the matching degree of general figure and intention text, and general match somebody with somebody to as much as possible for intention textual association
Figure.
It is apparent to those skilled in the art that, for convenience and simplicity of description, the system of the description,
The specific work process of device and unit, may be referred to the corresponding process in preceding method embodiment, will not be repeated here.
, can be by it in several embodiments provided herein, it should be understood that disclosed method and apparatus
Its mode is realized.For example, device embodiment described above is only schematical, for example, the division of the unit, only
Only a kind of division of logic function, can there is other dividing mode when actually realizing, such as multiple units or component can be tied
Another system is closed or is desirably integrated into, or some features can be ignored, or do not perform.It is another, it is shown or discussed
Coupling each other or direct-coupling or communication connection can be the INDIRECT COUPLINGs or logical of device or unit by some interfaces
Letter connection, can be electrical, machinery or other forms.
The unit illustrated as separating component can be or may not be it is physically separate, it is aobvious as unit
The part shown can be or may not be physical location, you can with positioned at a place, or can also be distributed to multiple
On NE.Some or all of unit therein can be selected to realize the mesh of this embodiment scheme according to the actual needs
's.
In addition, each functional unit in the application each embodiment can be integrated in a processing unit, can also
That unit is individually physically present, can also two or more units it is integrated in a unit.The integrated list
Member can both be realized in the form of hardware, it would however also be possible to employ hardware adds the form of SFU software functional unit to realize.
Fig. 3 shows the frame suitable for being used for the exemplary computer system/server 012 for realizing embodiment of the present invention
Figure.The computer system/server 012 that Fig. 3 is shown is only an example, to the function of the embodiment of the present invention and should not be used
Range band carrys out any limitation.
As shown in figure 3, computer system/server 012 is showed in the form of universal computing device.Computer system/clothes
The component of business device 012 can include but is not limited to：One or more processor or processing unit 016, system storage
028, the bus 018 of connection different system component (including system storage 028 and processing unit 016).
Bus 018 represents the one or more in a few class bus structures, including memory bus or Memory Controller,
Peripheral bus, graphics acceleration port, processor or the local bus using any bus structures in a variety of bus structures.Lift
For example, these architectures include but is not limited to industry standard architecture (ISA) bus, MCA (MAC)
Bus, enhanced isa bus, VESA's (VESA) local bus and periphery component interconnection (PCI) bus.
Computer system/server 012 typically comprises various computing systems computer-readable recording medium.These media can be appointed
The usable medium what can be accessed by computer system/server 012, including volatibility and non-volatile media, movably
With immovable medium.
System storage 028 can include the computer system readable media of form of volatile memory, for example, deposit at random
Access to memory (RAM) 030 and/or cache memory 032.Computer system/server 012 may further include other
Removable/nonremovable, volatile/non-volatile computer system storage medium.Only as an example, storage system 034 can
For reading and writing immovable, non-volatile magnetic media (Fig. 3 is not shown, is commonly referred to as " hard disk drive ").Although in Fig. 3
It is not shown, it can provide for the disc driver to may move non-volatile magnetic disk (such as " floppy disk ") read-write, and pair can
The CD drive of mobile anonvolatile optical disk (such as CD-ROM, DVD-ROM or other optical mediums) read-write.In these situations
Under, each driver can be connected by one or more data media interfaces with bus 018.Memory 028 can include
At least one program product, the program product has one group of (for example, at least one) program module, and these program modules are configured
To perform the function of various embodiments of the present invention.
Program/utility 040 with one group of (at least one) program module 042, can be stored in such as memory
In 028, such program module 042 includes --- but being not limited to --- operating system, one or more application program, other
The realization of network environment is potentially included in each or certain combination in program module and routine data, these examples.Journey
Sequence module 042 generally performs function and/or method in embodiment described in the invention.
Computer system/server 012 can also with one or more external equipments 014 (for example keyboard, sensing equipment,
Display 024 etc.) communication, in the present invention, computer system/server 012 is communicated with outside radar equipment, can also be with
One or more enables a user to the equipment communication interacted with the computer system/server 012, and/or with causing the meter
Any equipment (such as network interface card, modulation that calculation machine systems/servers 012 can be communicated with one or more of the other computing device
Demodulator etc.) communication.This communication can be carried out by input/output (I/O) interface 022.Also, computer system/clothes
Business device 012 can also pass through network adapter 020 and one or more network (such as LAN (LAN), wide area network (WAN)
And/or public network, such as internet) communication.As illustrated, network adapter 020 by bus 018 and computer system/
Other modules communication of server 012.Although it should be understood that not shown in Fig. 3, computer system/server 012 can be combined
Using other hardware and/or software module, include but is not limited to：Microcode, device driver, redundant processing unit, outside magnetic
Dish driving array, RAID system, tape drive and data backup storage system etc..
Processing unit 016 is stored in the program in system storage 028 by operation, so as to perform described in the invention
Function and/or method in embodiment.
Above-mentioned computer program can be arranged in computer-readable storage medium, i.e., the computer-readable storage medium is encoded with
Computer program, the program by one or more computers when being performed so that one or more computers are performed in the present invention
State the method flow shown in embodiment and/or device operation.
Over time, the development of technology, medium implication is more and more extensive, and the route of transmission of computer program is no longer limited by
Tangible medium, directly can also be downloaded from network etc..Any combination of one or more computer-readable media can be used.
Computer-readable medium can be computer-readable signal media or computer-readable recording medium.Computer-readable storage medium
Matter for example may be-but not limited to-system, device or the device of electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor, or
Combination more than person is any.The more specifically example (non exhaustive list) of computer-readable recording medium includes：With one
Or the electrical connections of multiple wires, portable computer diskette, hard disk, random access memory (RAM), read-only storage (ROM),
Erasable programmable read only memory (EPROM or flash memory), optical fiber, portable compact disc read-only storage (CD-ROM), light
Memory device, magnetic memory device or above-mentioned any appropriate combination.In this document, computer-readable recording medium can
Be it is any include or storage program tangible medium, the program can be commanded execution system, device or device use or
Person is in connection.
Computer-readable signal media can be included in a base band or as the data-signal of carrier wave part propagation,
Wherein carry computer-readable program code.The data-signal of this propagation can take various forms, including --- but
It is not limited to --- electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also be
Any computer-readable medium beyond computer-readable recording medium, the computer-readable medium can send, propagate or
Transmit for being used or program in connection by instruction execution system, device or device.
The program code included on computer-readable medium can be transmitted with any appropriate medium, including --- but do not limit
In --- wireless, electric wire, optical cable, RF etc., or above-mentioned any appropriate combination.
It can be write with one or more programming languages or its combination for performing the computer that the present invention is operated
Program code, described program design language includes object oriented program language-such as Java, Smalltalk, C++,
Also including conventional procedural programming language-such as " C " language or similar programming language.Program code can be with
Fully perform, partly perform on the user computer on the user computer, as independent software kit execution, a portion
Divide part execution or the execution completely on remote computer or server on the remote computer on the user computer.
Be related in the situation of remote computer, remote computer can be by the network of any kind --- including LAN (LAN) or
Wide area network (WAN) is connected to subscriber computer, or, it may be connected to outer computer (is for example provided using Internet service
Business comes by Internet connection).
Finally it should be noted that：Above example is only to the technical scheme for illustrating the application, rather than its limitations；Although
The application is described in detail with reference to the foregoing embodiments, it will be understood by those within the art that：It still may be used
To be modified to the technical scheme described in foregoing embodiments, or equivalent substitution is carried out to which part technical characteristic；
And these modification or replace, do not make appropriate technical solution essence depart from each embodiment technical scheme of the application spirit and
Scope.
Claims (14)
1. a kind of intelligence of internet promotion message matches somebody with somebody drawing method, it is characterised in that including：
Obtain the general of generic text and match somebody with somebody atlas；
The generic text is expanded into intention text；
According to the correlation with the intention text, concentrated from the general figure obtain the general figure of one or more with it is described
Intention text is combined, and generates internet promotion message.
2. according to the method described in claim 1, it is characterised in that the general of generic text that obtain matches somebody with somebody atlas, including：
Search daily record to the whole network user is analyzed and excavated, and obtains the interest main body and/or high frequency query word of the whole network user,
Clustered and duplicate removal processing, obtain candidate's text；
Scan for, obtain in the index of the picture text message pre-established using candidate's text as search keyword
With the index of candidate's text matches；
Corresponding with the index of candidate's text matches picture is obtained, using candidate's text as generic text, and by institute
Picture is stated as general figure；
All general figures are combined as general with atlas.
3. according to the method described in claim 1, it is characterised in that the general of generic text that obtain matches somebody with somebody atlas, including：
Advertising Law is not met wherein by artificial/automatic examination ＆ verification filtering and relates to yellow, gambling, the picture containing watermark.
4. according to the method described in claim 1, it is characterised in that described that generic text is expanded into intention text, including：
Generic text to be extended, the generation general text to be extended are inputted in the generation trigger model that training in advance is obtained
This corresponding intention text.
5. method according to claim 4, it is characterised in that generation trigger model training in advance in the following ways
Obtain：
Training sample is chosen, the training sample includes the positive and negative intention text of generic text and generic text；
Obtain training sample to input into original deep-neural-network, and the instruction is based on according to the original deep-neural-network
Practice the output result of sample, along the direction for minimizing loss function, reversely successively update each layer in the original deep layer network
Weighting parameters, to cause loss function minimum.
6. according to the method described in claim 1, it is characterised in that the basis and the correlation of the intention text, from institute
Stating general figure concentrates the selection general figure of one or more to be combined with the intention text, and letter is promoted in generation internet
Breath, including：
Extraction of semantics is carried out to the general figure, the picture feature of the general figure is obtained；
Picture and text feature extraction is carried out to generic text and the picture feature of the general figure；
The correlation of the general figure and intention text is obtained according to the picture and text feature and is ranked up according to correlation；
The forward general figure of one or more that will sort is combined with the intention text, generates internet promotion message.
7. a kind of intelligence of internet promotion message matches somebody with somebody map device, including acquiring unit, expanding element and generation unit, wherein,
The acquiring unit, matches somebody with somebody atlas for obtaining the general of generic text；
The expanding element, for generic text to be expanded into intention text；
The generation unit, for according to and the intention text correlation, from the general figure concentrate one width of selection or
Several general figures are combined with the intention text, generate internet promotion message.
8. device according to claim 7, it is characterised in that the acquiring unit includes following subelement：
Candidate's text obtains subelement, is analyzed and is excavated for the search daily record to the whole network user, obtains the whole network user's
Interest main body and/or high frequency query word, are clustered and duplicate removal processing, obtain candidate's text；
Index obtain subelement, for using candidate's text as search keyword in the picture text message pre-established
Scanned in index, obtain the index with candidate's text matches；
General figure obtains subelement, for obtaining picture corresponding with the index of candidate's text matches, by the candidate
Text is used as general figure as generic text, and using the picture；
It is general to combine subelement with atlas, it is general with atlas for all general figures to be combined as.
9. device according to claim 7, it is characterised in that the acquiring unit also includes examination ＆ verification subelement,
Picture for being concentrated to general figure carries out artificial/automatic examination ＆ verification, and filtering does not meet Advertising Law wherein and relates to yellow, gambling
The rich, picture containing watermark.
10. device according to claim 7, it is characterised in that the expanding element includes intention text and extends subelement,
For,
Generic text to be extended, the generation general text to be extended are inputted in the generation trigger model that training in advance is obtained
This corresponding intention text.
11. device according to claim 10, it is characterised in that the expanding element is trained including production trigger model
Subelement, is used for,
Training sample is chosen, the training sample includes the positive and negative intention text of generic text and generic text；
Obtain training sample to input into original deep-neural-network, and the instruction is based on according to the original deep-neural-network
Practice the output result of sample, along the direction for minimizing loss function, reversely successively update each layer in the original deep layer network
Weighting parameters, to cause loss function minimum.
12. device according to claim 7, it is characterised in that the generation unit, including following subelement：
Extraction of semantics subelement, for carrying out extraction of semantics according to the general figure using picture and text relationship type model；
Picture and text feature extraction subelement, picture and text feature extraction is carried out to generic text and the picture feature of the general figure；
Correlation obtains subelement, for obtaining the correlation of the general figure and intention text according to the picture and text feature simultaneously
It is ranked up according to correlation；
Internet promotion message generation subelement, for by the forward general figure of one or more and the intention text of sorting
It is combined, generates internet promotion message.
13. a kind of equipment, it is characterised in that the equipment includes：
One or more processors；
Storage device, for storing one or more programs,
When one or more of programs are by one or more of computing devices so that one or more of processors are real
The existing method as described in any in claim 1-6.
14. a kind of computer-readable recording medium, is stored thereon with computer program, it is characterised in that the program is by processor
The method as described in any in claim 1-6 is realized during execution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710374160.6A CN107193962B (en) | 2017-05-24 | 2017-05-24 | Intelligent map matching method and device for Internet promotion information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710374160.6A CN107193962B (en) | 2017-05-24 | 2017-05-24 | Intelligent map matching method and device for Internet promotion information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107193962A true CN107193962A (en) | 2017-09-22 |
| CN107193962B CN107193962B (en) | 2021-06-11 |
Family
ID=59874421
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710374160.6A Active CN107193962B (en) | 2017-05-24 | 2017-05-24 | Intelligent map matching method and device for Internet promotion information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107193962B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107948730A (en) * | 2017-10-30 | 2018-04-20 | 百度在线网络技术（北京）有限公司 | Method, apparatus, equipment and storage medium based on picture generation video |
| CN109543058A (en) * | 2018-11-23 | 2019-03-29 | 连尚（新昌）网络科技有限公司 | For the method for detection image, electronic equipment and computer-readable medium |
| CN109933802A (en) * | 2019-03-25 | 2019-06-25 | 腾讯科技（深圳）有限公司 | Picture and text matching process, device and storage medium |
| CN109948019A (en) * | 2019-01-10 | 2019-06-28 | 中央财经大学 | A kind of deep layer Network Data Capture method |
| CN109978581A (en) * | 2017-12-28 | 2019-07-05 | 百度在线网络技术（北京）有限公司 | Information launches optimization method, device, computer equipment and storage medium |
| CN110119506A (en) * | 2018-02-05 | 2019-08-13 | 北京京东尚科信息技术有限公司 | Text Extraction and system, equipment and storage medium |
| CN110211015A (en) * | 2018-02-28 | 2019-09-06 | 佛山科学技术学院 | A kind of water mark method based on feature object protection |
| CN110533743A (en) * | 2018-05-25 | 2019-12-03 | 阿里巴巴集团控股有限公司 | Processing method, equipment and the system of picture |
| CN110688508A (en) * | 2019-09-03 | 2020-01-14 | 北京字节跳动网络技术有限公司 | Image-text data expansion method and device and electronic equipment |
| WO2020103899A1 (en) * | 2018-11-23 | 2020-05-28 | 连尚（新昌）网络科技有限公司 | Method for generating inforgraphic information and method for generating image database |
| CN111311694A (en) * | 2018-12-11 | 2020-06-19 | 西安海平方网络科技有限公司 | Material arrangement method and device, computer equipment and storage medium |
| CN111581510A (en) * | 2020-05-07 | 2020-08-25 | 腾讯科技（深圳）有限公司 | Shared content processing method and device, computer equipment and storage medium |
| CN112150174A (en) * | 2019-06-27 | 2020-12-29 | 百度在线网络技术（北京）有限公司 | Advertisement matching method and device and electronic equipment |
| CN113377970A (en) * | 2020-03-10 | 2021-09-10 | 阿里巴巴集团控股有限公司 | Information processing method and device |
| CN115269844A (en) * | 2022-08-01 | 2022-11-01 | 腾讯科技（深圳）有限公司 | Model processing method and device, electronic equipment and storage medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080082497A1 (en) * | 2006-09-29 | 2008-04-03 | Leblang Jonathan A | Method and system for identifying and displaying images in response to search queries |
| CN101539916A (en) * | 2008-03-17 | 2009-09-23 | 亿维讯软件（北京）有限公司 | Initial patent retrieving device, secondary patent retrieving device and patent retrieving system |
| CN103226597A (en) * | 2013-04-19 | 2013-07-31 | 北京集奥聚合科技有限公司 | Keyword advertisement matching method based on natural semantics |
| CN103886063A (en) * | 2014-03-18 | 2014-06-25 | 国家电网公司 | Text retrieval method and device |
| CN104298786A (en) * | 2014-11-12 | 2015-01-21 | 广州出益信息科技有限公司 | Image retrieval method and device |
| CN104504109A (en) * | 2014-12-30 | 2015-04-08 | 百度在线网络技术（北京）有限公司 | Image search method and device |
| CN106021364A (en) * | 2016-05-10 | 2016-10-12 | 百度在线网络技术（北京）有限公司 | Method and device for establishing picture search correlation prediction model, and picture search method and device |
| CN106294584A (en) * | 2016-07-28 | 2017-01-04 | 北京百度网讯科技有限公司 | The training method of order models and device |
| CN106528588A (en) * | 2016-09-14 | 2017-03-22 | 厦门幻世网络科技有限公司 | Method and apparatus for matching resources for text information |
-
2017
- 2017-05-24 CN CN201710374160.6A patent/CN107193962B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080082497A1 (en) * | 2006-09-29 | 2008-04-03 | Leblang Jonathan A | Method and system for identifying and displaying images in response to search queries |
| CN101539916A (en) * | 2008-03-17 | 2009-09-23 | 亿维讯软件（北京）有限公司 | Initial patent retrieving device, secondary patent retrieving device and patent retrieving system |
| CN103226597A (en) * | 2013-04-19 | 2013-07-31 | 北京集奥聚合科技有限公司 | Keyword advertisement matching method based on natural semantics |
| CN103886063A (en) * | 2014-03-18 | 2014-06-25 | 国家电网公司 | Text retrieval method and device |
| CN104298786A (en) * | 2014-11-12 | 2015-01-21 | 广州出益信息科技有限公司 | Image retrieval method and device |
| CN104504109A (en) * | 2014-12-30 | 2015-04-08 | 百度在线网络技术（北京）有限公司 | Image search method and device |
| CN106021364A (en) * | 2016-05-10 | 2016-10-12 | 百度在线网络技术（北京）有限公司 | Method and device for establishing picture search correlation prediction model, and picture search method and device |
| CN106294584A (en) * | 2016-07-28 | 2017-01-04 | 北京百度网讯科技有限公司 | The training method of order models and device |
| CN106528588A (en) * | 2016-09-14 | 2017-03-22 | 厦门幻世网络科技有限公司 | Method and apparatus for matching resources for text information |
Non-Patent Citations (2)
| Title |
|---|
| 林子雨 等: ""基于关系数据库的关键词查询"", 《软件学报》 * |
| 郭永菊: "《电子器件领域专利检索策略及应用》", 30 June 2015, 知识产权出版社 * |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107948730B (en) * | 2017-10-30 | 2020-11-20 | 百度在线网络技术（北京）有限公司 | Method, device and equipment for generating video based on picture and storage medium |
| CN107948730A (en) * | 2017-10-30 | 2018-04-20 | 百度在线网络技术（北京）有限公司 | Method, apparatus, equipment and storage medium based on picture generation video |
| CN109978581A (en) * | 2017-12-28 | 2019-07-05 | 百度在线网络技术（北京）有限公司 | Information launches optimization method, device, computer equipment and storage medium |
| CN110119506A (en) * | 2018-02-05 | 2019-08-13 | 北京京东尚科信息技术有限公司 | Text Extraction and system, equipment and storage medium |
| CN110119506B (en) * | 2018-02-05 | 2023-12-08 | 北京京东尚科信息技术有限公司 | Text extraction method, system, equipment and storage medium |
| CN110211015A (en) * | 2018-02-28 | 2019-09-06 | 佛山科学技术学院 | A kind of water mark method based on feature object protection |
| CN110211015B (en) * | 2018-02-28 | 2022-12-20 | 佛山科学技术学院 | Watermark method based on characteristic object protection |
| CN110533743A (en) * | 2018-05-25 | 2019-12-03 | 阿里巴巴集团控股有限公司 | Processing method, equipment and the system of picture |
| CN110533743B (en) * | 2018-05-25 | 2023-03-21 | 阿里巴巴集团控股有限公司 | Picture processing method, device and system |
| CN109543058A (en) * | 2018-11-23 | 2019-03-29 | 连尚（新昌）网络科技有限公司 | For the method for detection image, electronic equipment and computer-readable medium |
| WO2020103899A1 (en) * | 2018-11-23 | 2020-05-28 | 连尚（新昌）网络科技有限公司 | Method for generating inforgraphic information and method for generating image database |
| CN111311694A (en) * | 2018-12-11 | 2020-06-19 | 西安海平方网络科技有限公司 | Material arrangement method and device, computer equipment and storage medium |
| CN111311694B (en) * | 2018-12-11 | 2023-04-18 | 西安海平方网络科技有限公司 | Material arrangement method and device, computer equipment and storage medium |
| CN109948019A (en) * | 2019-01-10 | 2019-06-28 | 中央财经大学 | A kind of deep layer Network Data Capture method |
| CN109948019B (en) * | 2019-01-10 | 2021-10-08 | 中央财经大学 | Deep network data acquisition method |
| CN109933802A (en) * | 2019-03-25 | 2019-06-25 | 腾讯科技（深圳）有限公司 | Picture and text matching process, device and storage medium |
| CN109933802B (en) * | 2019-03-25 | 2023-05-26 | 腾讯科技（深圳）有限公司 | Image-text matching method, image-text matching device and storage medium |
| CN112150174A (en) * | 2019-06-27 | 2020-12-29 | 百度在线网络技术（北京）有限公司 | Advertisement matching method and device and electronic equipment |
| CN112150174B (en) * | 2019-06-27 | 2024-04-02 | 百度在线网络技术（北京）有限公司 | Advertisement picture allocation method and device and electronic equipment |
| CN110688508A (en) * | 2019-09-03 | 2020-01-14 | 北京字节跳动网络技术有限公司 | Image-text data expansion method and device and electronic equipment |
| CN113377970A (en) * | 2020-03-10 | 2021-09-10 | 阿里巴巴集团控股有限公司 | Information processing method and device |
| CN111581510A (en) * | 2020-05-07 | 2020-08-25 | 腾讯科技（深圳）有限公司 | Shared content processing method and device, computer equipment and storage medium |
| CN111581510B (en) * | 2020-05-07 | 2024-02-09 | 腾讯科技（深圳）有限公司 | Shared content processing method, device, computer equipment and storage medium |
| CN115269844A (en) * | 2022-08-01 | 2022-11-01 | 腾讯科技（深圳）有限公司 | Model processing method and device, electronic equipment and storage medium |
| CN115269844B (en) * | 2022-08-01 | 2024-03-29 | 腾讯科技（深圳）有限公司 | Model processing method, device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107193962B (en) | 2021-06-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107193962A (en) | A kind of intelligent figure method and device of internet promotion message | |
| CN107832468B (en) | Demand recognition methods and device | |
| CN107832432A (en) | A kind of search result ordering method, device, server and storage medium | |
| US20180247405A1 (en) | Automatic detection and semantic description of lesions using a convolutional neural network | |
| EP3711002A1 (en) | Resource-efficient generation of a knowledge graph | |
| CN104572072B (en) | A kind of language transfer method and equipment to the program based on MVC pattern | |
| CN111026842A (en) | Natural language processing method, natural language processing device and intelligent question-answering system | |
| CN107491432A (en) | Low quality article recognition methods and device, equipment and medium based on artificial intelligence | |
| CN107423398A (en) | Exchange method, device, storage medium and computer equipment | |
| CN106105096A (en) | System and method for continuous social communication | |
| CN108171260A (en) | A kind of image identification method and system | |
| CN109543549A (en) | Image processing method and device, mobile end equipment, server for more people's Attitude estimations | |
| CN108121800A (en) | Information generating method and device based on artificial intelligence | |
| CN110032630A (en) | Talk about art recommendation apparatus, method and model training equipment | |
| CN107301170A (en) | The method and apparatus of cutting sentence based on artificial intelligence | |
| CN110490625A (en) | User preference determines method and device, electronic equipment, storage medium | |
| CN108491421A (en) | A kind of method, apparatus, equipment and computer storage media generating question and answer | |
| CN109726661A (en) | Image processing method and device, medium and calculating equipment | |
| CN109783624A (en) | Answer generation method, device and the intelligent conversational system in knowledge based library | |
| CN112528639B (en) | Object recognition method and device, storage medium and electronic equipment | |
| CN110110663A (en) | A kind of age recognition methods and system based on face character | |
| CN110019519A (en) | Data processing method, device, storage medium and electronic device | |
| CN106354856A (en) | Enhanced deep neural network search method and device based on artificial intelligence | |
| CN109992781A (en) | Processing, device, storage medium and the processor of text feature | |
| CN109214417A (en) | The method for digging and device, computer equipment and readable medium that user is intended to |
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 |