CN106095983B - A kind of similarity based on personalized deep neural network determines method and device - Google Patents
A kind of similarity based on personalized deep neural network determines method and device Download PDFInfo
- Publication number
- CN106095983B CN106095983B CN201610445828.7A CN201610445828A CN106095983B CN 106095983 B CN106095983 B CN 106095983B CN 201610445828 A CN201610445828 A CN 201610445828A CN 106095983 B CN106095983 B CN 106095983B
- Authority
- CN
- China
- Prior art keywords
- vector
- user
- similarity
- neural network
- query
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention discloses a kind of similarities based on personalized deep neural network to determine method and device.This method comprises: obtain the query text and user personalized information of user's input, wherein the attribute information for the intelligent terminal that the user personalized information is historical search behavior according to the user or the user holds determines;Deep neural network processing is carried out to the query text, search entry and the user personalized information, and determines the similarity between the query text and described search entry according to deep neural network processing result.The technical solution of the embodiment of the present invention improves the modelling effect that traditional semantic similarity determines, to improve the accuracy of similarity between query text and search entry by having merged user personalized information in deep neural network learns.
Description
Technical field
The present invention relates to fields of communication technology more particularly to a kind of similarity based on personalized deep neural network to determine
A kind of image partition method of method and device and device.
Background technique
For the query text of user's input, the basis that search result is search engine system is returned to for user.Wherein really
Determine the similarity between the query text search entry corresponding with query text of user's input, is to return to search result for user
Premise.
Currently, determine method based on the similarity of deep neural network (Deep Neural Network, DNN), only to
The query text and the corresponding search entry of query text of family input carry out DNN processing, and obtain phase based on DNN processing result
Like degree, and the polysemy or the more word phenomenons of justice of natural language text are not considered, thus the accuracy that existing similarity determines
It is low, it is not able to satisfy the individual demand of user, tends not to reach satisfactory effect in practical applications.
Summary of the invention
In view of this, the embodiment of the present invention provide a kind of similarity based on personalized deep neural network determine method and
Device, to improve the accuracy that similarity determines, to meet the individual demand of user.
In a first aspect, the embodiment of the invention provides a kind of similarity determination sides based on personalized deep neural network
Method, comprising:
The query text and user personalized information for obtaining user's input, wherein the user personalized information is foundation
What the attribute information for the intelligent terminal that the historical search behavior of the user or the user hold determined;
Deep neural network processing is carried out to the query text, search entry and the user personalized information, and
The similarity between the query text and described search entry is determined according to deep neural network processing result.
Second aspect, the embodiment of the invention provides a kind of similarity determining devices based on personalized deep neural network
Side, comprising:
Query text obtains module, for obtaining the query text and user personalized information of user's input, wherein institute
State the attribute for the intelligent terminal that user personalized information is historical search behavior according to the user or the user holds
What information determined;
Similarity determining module, for being carried out to the query text, search entry and the user personalized information
Deep neural network processing, and determined between the query text and described search entry according to deep neural network processing result
Similarity.
Technical solution provided in an embodiment of the present invention, by having merged user individual in deep neural network learns
Information, so that query text, the corresponding search entry of query text and the user personalized information according to user's input obtain
Personalized neural network improves the modelling effect that traditional semantic similarity determines, to improve query text and search
The accuracy of similarity between rope entry.
Detailed description of the invention
Fig. 1 is that a kind of similarity based on personalized deep neural network that the embodiment of the present invention one provides determines method
Flow chart;
Fig. 2 a is that a kind of similarity based on personalized deep neural network provided by Embodiment 2 of the present invention determines method
Flow chart;
Fig. 2 b is the original that the similarity provided by Embodiment 2 of the present invention based on personalized deep neural network determines method
Manage schematic diagram;
Fig. 3 a is that a kind of similarity based on personalized deep neural network that the embodiment of the present invention three provides determines method
Flow chart;
Fig. 3 b is the original that the similarity based on personalized deep neural network that the embodiment of the present invention three provides determines method
Manage schematic diagram;
Fig. 4 a is that a kind of similarity based on personalized deep neural network that the embodiment of the present invention four provides determines method
Flow chart;
Fig. 4 b is the original that the similarity based on personalized deep neural network that the embodiment of the present invention four provides determines method
Manage schematic diagram;
Fig. 5 is a kind of similarity determining device based on personalized deep neural network that the embodiment of the present invention five provides
Structure chart.
Specific embodiment
The present invention is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining the present invention rather than limiting the invention.It also should be noted that in order to just
Only the parts related to the present invention are shown in description, attached drawing rather than entire infrastructure.
Embodiment one
Fig. 1 is that a kind of similarity based on personalized deep neural network that the embodiment of the present invention one provides determines method
Flow chart.The method of the present embodiment can be by being executed based on the similarity determining device of personalized deep neural network, the dress
Setting can be realized by way of hardware and/or software, and the method for the present embodiment is generally applicable to user and wants to be inquired
The situation of similarity between text and search entry.It is provided in this embodiment based on personalized deep neural network referring to Fig. 1
It is as follows that similarity determines that method can specifically include:
S11, the query text and user personalized information for obtaining user's input.
Specifically, detecting user, by input equipments such as touch screen or keyboards, the progress text in search input frame is defeated
It is fashionable, obtain the query text and user personalized information of user's input.Wherein, user personalized information can be according to institute
State the historical search behavior determination of user.Correspondingly, user can also be obtained and log in user identifier used in search engine
Information such as User Identity number (Identification, ID) etc., and user individual is obtained according to user identity information
Information.The historical search behavior of user may include user historical query text and user it is corresponding from historical query text
The historical search entry selected in search entry.Wherein, user personalized information is also possible to the intelligence held according to the user
The attribute information of energy terminal determines.
Illustratively, user personalized information may include the interest of user, region where user, alternatively, user holds
Model, brand, operating system or the browser of intelligent terminal etc..Wherein, the interest of user may include sport, literature, IT,
Game, electronics, fruit, food or agricultural product etc. are that the historical search behavior of foundation user obtains.Region where user can also
To be obtained according to user's history search behavior.
S12, the query text, search entry and the user personalized information are carried out at deep neural network
Reason, and the similarity between the query text and described search entry is determined according to deep neural network processing result.
Wherein, the vector that deep neural network technology is used to learn word indicates, i.e., is handled by deep neural network by word
It is expressed as the vector of real number composition.Specifically, using deep neural network learning art, according to user personalized information to user
Individual is modeled, and obtains user's insertion of the corresponding personalization of each user, and user's insertion is introduced into conventional depth mind
Through in network semantic model, and based on the deep neural network semantic model being embedded in comprising user, determines query text and search
Similarity between rope entry, such as cosine similarity.Since the deep neural network semantic model being embedded in comprising user had both been kept
Semantic similarity between traditional query text and search entry, and there is personalized user's insertion, with utmostly
Ground meets the individual demand of user, improves the accuracy of similarity between query text and search entry.
It should be noted that search entry refers to the corresponding search entry of query text, and to search in the present embodiment
Entry number is not especially limited.For example, search entry can be agricultural product price and mobile phone zero is matched when query text is apple
Part etc..
Technical solution provided in this embodiment, by having merged user personalized information in deep neural network learns,
To obtain individual character according to query text, the corresponding search entry of query text and the user personalized information of user's input
The neural network of change improves the modelling effect that traditional semantic similarity determines, to improve query text and search entry
Between similarity accuracy.
Illustratively, it is determined between the query text and described search entry according to deep neural network processing result
It may include: according to the similarity between the query text and described search entry, to described search entry after similarity
It is ranked up.Specifically, being directed to any search entry, if the similarity of the search entry and query text is bigger, the search
Entry sequence is more preceding.
Embodiment two
The present embodiment provides a kind of new based on personalized deep neural network on the basis of the above embodiment 1
Similarity determines method.User personalized information is considered as to a part of information of query text in the present embodiment, in depth mind
The expression of user personalized information is merged with the expression of query text through network model top layer.
Fig. 2 a is that a kind of similarity based on personalized deep neural network provided by Embodiment 2 of the present invention determines method
Flow chart, Fig. 2 b is the original that the similarity provided by Embodiment 2 of the present invention based on personalized deep neural network determines method
Manage schematic diagram.In order to make it easy to understand, with query text be apple in the present embodiment, search entry is carried out for agricultural product price
Explanation.In conjunction with Fig. 2 a and Fig. 2 b, the similarity provided in this embodiment based on personalized deep neural network determines that method is specific
May include as follows:
S21, the query text and user personalized information for obtaining user's input.
Wherein, the user personalized information is that historical search behavior or the user according to the user are held
Intelligent terminal attribute information determine.Specifically, obtaining query text-apple that user inputs in search input frame.
S22, it is handled by deep neural network, the query text is expressed as query vector, search entry is indicated
At locating vector, and the user personalized information is expressed as user and is embedded in vector.
Specifically, be expressed as query text by word insertion (Word Embedding) processing to inquire intermediate vector, and
It is handled by deep neural network, inquiry intermediate vector is expressed as query vector.Correspondingly, obtain that query text is corresponding searches
Search entry is expressed as searching for intermediate vector by word insertion processing, and passes through depth nerve net by rope entry-agricultural product price
Search intermediate vector is expressed as query vector by network processing;User personalized information is obtained, it is by word insertion processing that user is a
Property information processing at user's intermediate vector, and by deep neural network processing user's intermediate vector is expressed as user's insertion
Vector.
S23, the fusion query vector and the user are embedded in vector, to obtain new query vector.
Specifically, can determine that the weighted value of query vector and user are embedded in the weighted value of vector, and the two weighted value
The sum of be equal to 1, and based on both weighted value query vector and user insertion vector are weighted are superimposed, to obtain new inquiry
Vector.It should be noted that the weighted value of general inquiry vector is greater than or equal to the weighted value that user is embedded in vector, such as the two power
Weight values can be equal to 0.5.
S24, similarity between new query vector and described search vector is determined.
Specifically, new query vector and locating vector can be plotted in vector space, new query vector is determined
Angle between locating vector, and the corresponding cosine value of angle is obtained, which is used to characterize the cosine similarity of vector.
Angle is smaller, and cosine value is closer to 1, then new query vector is more similar to locating vector.
Technical solution provided in this embodiment, by the way that user personalized information to be considered as to a part of information of query text,
The expression of user personalized information is merged with the expression of query text in deep neural network model top layer, is obtained new
Query vector is handled by deep neural network search entry being expressed as locating vector, and by new query vector and search
Similarity between vector is as the similarity between query text and search entry.Since this method is in deep neural network
In habit, user personalized information has been merged, has improved the accuracy of similarity between query text and search entry.
Embodiment three
The present embodiment provides a kind of new based on personalized deep neural network on the basis of the above embodiment 1
Similarity determines method.The vector for increasing search entry in the present embodiment indicates to indicate with the vector of user personalized information
Between similarity determine.
Fig. 3 a is that a kind of similarity based on personalized deep neural network that the embodiment of the present invention three provides determines method
Flow chart, Fig. 3 b is that the similarity based on personalized deep neural network that the embodiment of the present invention three provides determines the original of method
Manage schematic diagram.In order to make it easy to understand, with query text be apple in the present embodiment, search entry is carried out for agricultural product price
Explanation.In conjunction with Fig. 3 a and Fig. 3 b, the similarity provided in this embodiment based on personalized deep neural network determines that method is specific
May include as follows:
S31, the query text and user personalized information for obtaining user's input.
Wherein, the user personalized information is that historical search behavior or the user according to the user are held
Intelligent terminal attribute information determine.
S32, it is handled by deep neural network, the query text is expressed as query vector, search entry is indicated
At locating vector, and the user personalized information is expressed as user and is embedded in vector.
Specifically, can be believed based on deep neural network semantic model query text, search entry and user individual
Breath is handled, and is obtained query vector, locating vector and user and is embedded in vector.
S33, the first similarity between the query vector and described search vector is determined.
Specifically, the cosine similarity between query vector and locating vector can be determined, as the first similarity.
S34, the second similarity that the user is embedded between vector and described search vector is determined.
Specifically, the cosine similarity that user is embedded between vector and locating vector can be determined, as the second similarity.
S35, according to first similarity and second similarity, determine the query text and described search entry
Between similarity.
Specifically, the weighted value of the first similarity and the weighted value of the second similarity can be determined, and the two weighted value
The sum of be equal to 1, and based on both weighted value the first similarity be weighted with the second similarity be superimposed, using stack result as
Similarity between query text and search entry.It should be noted that the weighted value of general first similarity is greater than or equal to
The weighted value of second similarity, such as weighted value of the first similarity are 0.7, and the weighted value of the second similarity is 0.3.
Technical solution provided in this embodiment, the vector by increasing search entry indicate with user personalized information to
Similarity between amount expression determines, to obtain the second similarity, and determines the first phase between query vector and locating vector
The similarity between query text and search entry is obtained like degree, and according to the second similarity and the first similarity.Due to the party
Method has merged user personalized information, has improved similar between query text and search entry in deep neural network study
The accuracy of degree.
Example IV
The present embodiment provides a kind of new based on personalized deep neural network on the basis of the above embodiment 1
Similarity determines method.In the present embodiment by user personalized information as a part of query text, in depth nerve net
Network model bottom merges user personalized information with query text, is equivalent in query text and increases user personality
Change information.
Fig. 4 a is that a kind of similarity based on personalized deep neural network that the embodiment of the present invention four provides determines method
Flow chart, Fig. 4 b is that the similarity based on personalized deep neural network that the embodiment of the present invention four provides determines the original of method
Manage schematic diagram.In order to make it easy to understand, with query text be apple in the present embodiment, search entry is carried out for agricultural product price
Explanation.In conjunction with Fig. 4 a and Fig. 4 b, the similarity provided in this embodiment based on personalized deep neural network determines that method is specific
May include as follows:
S41, the query text and user personalized information for obtaining user's input.
Wherein, the user personalized information is that historical search behavior or the user according to the user are held
Intelligent terminal attribute information determine.
S42, it is handled by word insertion, the query text is expressed as to inquire intermediate vector, and by the user personality
Change information and is expressed as user's intermediate vector.
S43, the fusion inquiry intermediate vector and user's intermediate vector increase vector newly to obtain user.
Illustratively, the inquiry intermediate vector and user's intermediate vector are merged, increases vector newly to obtain user, it can
To include: to handle by deep neural network, the query text is expressed as query vector, and the user individual is believed
Breath is expressed as user and is embedded in vector;It determines the third similarity between the query vector and described search vector, and determines institute
State the 4th similarity that user is embedded between vector and described search vector;To the third similarity and the 4th similarity
Make normalized, obtains the first weight of the inquiry intermediate vector and the second weight of user's intermediate vector;Foundation
First weight and second weight merge the inquiry intermediate vector and user's intermediate vector, to obtain user
Newly-increased vector.
It should be noted that can also need according to user, the first weight and user's intermediate vector of query vector are set
The second weight.
S44, it is handled by deep neural network, increases the user newly vector and be expressed as new query vector, and will search
Rope entry representation is at locating vector.
It should be noted that deep neural network Processing Algorithm is not especially limited in the present embodiment, it only need to be based on deep
Spend neural network semantic model, increase user newly vector and be expressed as new query vector, and by search entry be expressed as search to
Amount.
S45, according to the new query vector and described search vector, determine the query text and described search entry
Between similarity.
Technical solution provided in this embodiment, by a part by user personalized information as query text, in depth
Degree neural network model bottom merges user personalized information with query text, i.e., is used by increasing in query text
Family customized information improves the accuracy of similarity between query text and search entry.
Embodiment five
Fig. 5 is a kind of similarity determining device based on personalized deep neural network that the embodiment of the present invention five provides
Structure chart.The device is generally applicable to the situation that user wants to obtain similarity between query text and search entry.Referring to
The specific structure of Fig. 5, the similarity determining device provided in this embodiment based on personalized deep neural network are as follows:
Query text obtains module 51, for obtaining the query text and user personalized information of user's input, wherein
The category for the intelligent terminal that the user personalized information is historical search behavior according to the user or the user holds
Property information determine;
Similarity determining module 52, for the query text, search entry and the user personalized information into
The processing of row deep neural network, and according to deep neural network processing result determine the query text and described search entry it
Between similarity.
Illustratively, similarity determining module 52 may include:
First processing units will for by deep neural network processing, the query text to be expressed as query vector
Search entry is expressed as locating vector, and the user personalized information is expressed as user and is embedded in vector;
Primary vector integrated unit is embedded in vector for merging the query vector and the user, to obtain new look into
Ask vector;
First similarity determining unit, for determining the similarity between new query vector and described search vector.
Illustratively, similarity determining module 52 may include:
The second processing unit will for by deep neural network processing, the query text to be expressed as query vector
Search entry is expressed as locating vector, and the user personalized information is expressed as user and is embedded in vector;
Primary vector determination unit, for determining the first similarity between the query vector and described search vector;
Secondary vector determination unit, for determining that it is similar to second between described search vector that the user is embedded in vector
Degree;
Second similarity determining unit, for looking into described according to first similarity and second similarity, determining
Ask the similarity between text and described search entry.
Illustratively, similarity determining module 52 may include:
Word is embedded in processing unit, for handling by word insertion, the query text is expressed as to inquire intermediate vector, and
The user personalized information is expressed as user's intermediate vector;
Secondary vector integrated unit, for merging the inquiry intermediate vector and user's intermediate vector, to be used
Family increases vector newly;
Third processing unit increases the user newly vector and is expressed as new look into for being handled by deep neural network
Vector is ask, and search entry is expressed as locating vector;
Third similarity determining unit, for looking into described according to the new query vector and described search vector, determining
Ask the similarity between text and described search entry.
Exemplary, secondary vector integrated unit may include:
The query text is expressed as inquiring by Processing with Neural Network subelement for handling by deep neural network
Vector, and the user personalized information is expressed as user and is embedded in vector;
Vector determines subelement, for determining the third similarity between the query vector and described search vector, and
Determine the 4th similarity that the user is embedded between vector and described search vector;
Weight determines subelement, for making normalized to the third similarity and the 4th similarity, obtains
First weight of the inquiry intermediate vector and the second weight of user's intermediate vector;
Vector determines subelement, for according to first weight and second weight, merge among the inquiry to
Amount and user's intermediate vector increase vector newly to obtain user.
Illustratively, user personalized information includes the interest of user, region where user, alternatively, the intelligence that user holds
Model, brand, operating system or the browser of energy terminal.
Illustratively, above-mentioned apparatus may include:
Search entry sorting module, for according to deep neural network processing result determine the query text with it is described
After similarity between search entry, according to the similarity between the query text and described search entry, searched to described
Rope entry is ranked up.
In addition, in order to verify the validity of the deep neural network model provided in the present embodiment, for different indexs into
Following multiple groups comparative experiments is gone, experiment effect has been more than the optimum of impersonal theory deep neural network.
Experiment purpose: the effect that personalized neural network judges text similarity under true Webpage search environment is verified
Fruit.
Training data: the training set of the log corpus (in August, 2015 in October, -2015) from Mr. Yu's search engine.
Test data: the test set of the log corpus (in November, 2015) from Mr. Yu's search engine.
Appraisal procedure: the positive backward ratio and AUC (Area Under roc Curve, under ROC curve of test data are assessed
Area) index.
Experimental setup: scheme one is respectively adopted: in deep neural network model top layer by the expression of user personalized information
It is merged with the expression of query text, scheme two: increasing the vector of the vector expression and user personalized information of search entry
Similarity between expression determines, scheme three: in deep neural network model bottom by user personalized information and query text
It is merged and impersonal theory model (not considering user personalized information), on the training data training pattern, and is existed respectively
Positive backward ratio and AUC index are tested in test, last and non-personalized traditional neural network model compares, as a result such as
Under:
Relative to traditional neural network model, three kinds of personalized neural network schemes achieve excellent effect and mention
It rises, and it is best using user personalized information as adjunct word to be added to the scheme works in query text in scheme three, non-
Positive backward improves 0.377 (general positive backward is effect promoting than improving 0.2 in searching order on the basis of personalized model
Obviously), AUC index improves 3.6%.To sum up, what is provided in the embodiment of the present invention is looked into based on the determination of personalized deep neural network
The method for asking text and search entry similarity can significantly improve the effect that traditional semantic similarity determines.
Technical solution provided in this embodiment uses depth learning technology, avoids relative to traditional Personalized Policies
Relatively complicated Feature Engineering, reduces human cost, and do not need the higher user profile of maintenance cost
(profile), thus reduce cost and resource consumption.
Note that the above is only a better embodiment of the present invention and the applied technical principle.It will be appreciated by those skilled in the art that
The invention is not limited to the specific embodiments described herein, be able to carry out for a person skilled in the art it is various it is apparent variation,
It readjusts and substitutes without departing from protection scope of the present invention.Therefore, although being carried out by above embodiments to the present invention
It is described in further detail, but the present invention is not limited to the above embodiments only, without departing from the inventive concept, also
It may include more other equivalent embodiments, and the scope of the invention is determined by the scope of the appended claims.
Claims (8)
1. a kind of similarity based on personalized deep neural network determines method, comprising:
The query text and user personalized information of user's input are obtained, wherein the user personalized information is according to described in
What the attribute information for the intelligent terminal that the historical search behavior of user or the user hold determined;
Deep neural network processing, and foundation are carried out to the query text, search entry and the user personalized information
Deep neural network processing result determines the similarity between the query text and described search entry;
Wherein, deep neural network processing is carried out to the query text, search entry and the user personalized information, and
The similarity between the query text and described search entry is determined according to deep neural network processing result, comprising:
Handled by deep neural network, the query text be expressed as query vector, by search entry be expressed as search to
Amount, and the user personalized information is expressed as user and is embedded in vector;
It merges the query vector and the user is embedded in vector, to obtain new query vector;
Determine the similarity between new query vector and described search vector;
Alternatively, deep neural network processing is carried out to the query text, search entry and the user personalized information, and
The similarity between the query text and described search entry is determined according to deep neural network processing result, comprising:
It is handled by word insertion, the query text is expressed as to inquire intermediate vector, and by the user personalized information table
It is shown as user's intermediate vector;
The inquiry intermediate vector and user's intermediate vector are merged, increases vector newly to obtain user;
It is handled by deep neural network, increases the user newly vector and be expressed as new query vector, and by search entry table
It is shown as locating vector;
According to the new query vector and described search vector, the phase between the query text and described search entry is determined
Like degree.
2. the method according to claim 1, wherein merge among the inquiry intermediate vector and the user to
Amount increases vector newly to obtain user, comprising:
It is handled by deep neural network, the query text is expressed as query vector, and by the user personalized information
It is expressed as user and is embedded in vector;
It determines the third similarity between the query vector and described search vector, and determines that the user is embedded in vector and institute
State the 4th similarity between locating vector;
Normalized is made to the third similarity and the 4th similarity, obtains the first power of the inquiry intermediate vector
Second weight of weight and user's intermediate vector;
According to first weight and second weight, the inquiry intermediate vector and user's intermediate vector are merged, with
It obtains user and increases vector newly.
3. the method according to claim 1, wherein user personalized information includes the interest of user, Yong Husuo
In region, alternatively, the model for the intelligent terminal that user holds, brand, operating system or browser.
4. the method according to claim 1, wherein determining the inquiry according to deep neural network processing result
After similarity between text and described search entry, comprising:
According to the similarity between the query text and described search entry, described search entry is ranked up.
5. a kind of similarity determining device based on personalized deep neural network, comprising:
Query text obtains module, for obtaining the query text and user personalized information of user's input, wherein the use
The attribute information for the intelligent terminal that family customized information is historical search behavior according to the user or the user holds
Determining;Similarity determining module, it is deep for being carried out to the query text, search entry and the user personalized information
Processing with Neural Network is spent, and is determined between the query text and described search entry according to deep neural network processing result
Similarity;
Wherein, the similarity determining module includes:
The query text is expressed as query vector, will searched for by first processing units for being handled by deep neural network
The user personalized information is expressed as user and is embedded in vector by entry representation at locating vector;
Primary vector integrated unit is embedded in vector for merging the query vector and the user, with obtain new inquiry to
Amount;
First similarity determining unit, for determining the similarity between new query vector and described search vector;
Alternatively, similarity determining module includes:
Word is embedded in processing unit, for handling by word insertion, the query text is expressed as inquiry intermediate vector, and by institute
It states user personalized information and is expressed as user's intermediate vector;
Secondary vector integrated unit is new to obtain user for merging the inquiry intermediate vector and user's intermediate vector
Increase vector;
Third processing unit, for being handled by deep neural network, by the newly-increased vector of the user be expressed as new inquiry to
Amount, and search entry is expressed as locating vector;
Third similarity determining unit, for determining the inquiry text according to the new query vector and described search vector
Similarity between sheet and described search entry.
6. device according to claim 5, which is characterized in that secondary vector integrated unit includes:
Processing with Neural Network subelement, for by deep neural network processing, the query text to be expressed as query vector,
And the user personalized information is expressed as user and is embedded in vector;
Vector determines subelement, for determining the third similarity between the query vector and described search vector, and determines
The user is embedded in the 4th similarity between vector and described search vector;
Weight determines subelement, for making normalized to the third similarity and the 4th similarity, obtains described
Inquire the first weight of intermediate vector and the second weight of user's intermediate vector;
Vector determines subelement, for according to first weight and second weight, merge the inquiry intermediate vector and
User's intermediate vector increases vector newly to obtain user.
7. device according to claim 5, which is characterized in that user personalized information includes the interest of user, Yong Husuo
In region, alternatively, the model for the intelligent terminal that user holds, brand, operating system or browser.
8. device according to claim 5 characterized by comprising
Search entry sorting module, for determining the query text and described search according to deep neural network processing result
After similarity between entry, according to the similarity between the query text and described search entry, to described search item
Mesh is ranked up.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610445828.7A CN106095983B (en) | 2016-06-20 | 2016-06-20 | A kind of similarity based on personalized deep neural network determines method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610445828.7A CN106095983B (en) | 2016-06-20 | 2016-06-20 | A kind of similarity based on personalized deep neural network determines method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106095983A CN106095983A (en) | 2016-11-09 |
CN106095983B true CN106095983B (en) | 2019-11-26 |
Family
ID=57237186
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610445828.7A Active CN106095983B (en) | 2016-06-20 | 2016-06-20 | A kind of similarity based on personalized deep neural network determines method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106095983B (en) |
Families Citing this family (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107491518B (en) | 2017-08-15 | 2020-08-04 | 北京百度网讯科技有限公司 | Search recall method and device, server and storage medium |
CN110633407B (en) * | 2018-06-20 | 2022-05-24 | 百度在线网络技术(北京)有限公司 | Information retrieval method, device, equipment and computer readable medium |
RU2731658C2 (en) | 2018-06-21 | 2020-09-07 | Общество С Ограниченной Ответственностью "Яндекс" | Method and system of selection for ranking search results using machine learning algorithm |
RU2733481C2 (en) | 2018-12-13 | 2020-10-01 | Общество С Ограниченной Ответственностью "Яндекс" | Method and system for generating feature for ranging document |
CN110765368B (en) * | 2018-12-29 | 2020-10-27 | 滴图(北京)科技有限公司 | Artificial intelligence system and method for semantic retrieval |
RU2744029C1 (en) | 2018-12-29 | 2021-03-02 | Общество С Ограниченной Ответственностью "Яндекс" | System and method of forming training set for machine learning algorithm |
CN110162191A (en) * | 2019-04-03 | 2019-08-23 | 腾讯科技(深圳)有限公司 | A kind of expression recommended method, device and storage medium |
KR20210015524A (en) | 2019-08-02 | 2021-02-10 | 삼성전자주식회사 | Method and electronic device for quantifying user interest |
CN114666812A (en) * | 2020-12-24 | 2022-06-24 | 华为技术有限公司 | Information processing method and device |
CN113485801B (en) * | 2021-06-25 | 2023-07-28 | 中国科学技术大学苏州高等研究院 | Real-time DNN scheduling system and method based on neural network similarity modeling |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103020164A (en) * | 2012-11-26 | 2013-04-03 | 华北电力大学 | Semantic search method based on multi-semantic analysis and personalized sequencing |
CN103488664A (en) * | 2013-05-03 | 2014-01-01 | 中国传媒大学 | Image retrieval method |
CN104462357A (en) * | 2014-12-08 | 2015-03-25 | 百度在线网络技术(北京)有限公司 | Method and device for realizing personalized search |
CN104484380A (en) * | 2014-12-09 | 2015-04-01 | 百度在线网络技术(北京)有限公司 | Personalized search method and personalized search device |
CN104598611A (en) * | 2015-01-29 | 2015-05-06 | 百度在线网络技术(北京)有限公司 | Method and system for sequencing search entries |
CN104679771A (en) * | 2013-11-29 | 2015-06-03 | 阿里巴巴集团控股有限公司 | Individual data searching method and device |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104239458A (en) * | 2014-09-02 | 2014-12-24 | 百度在线网络技术(北京)有限公司 | Method and device for representing search results |
-
2016
- 2016-06-20 CN CN201610445828.7A patent/CN106095983B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103020164A (en) * | 2012-11-26 | 2013-04-03 | 华北电力大学 | Semantic search method based on multi-semantic analysis and personalized sequencing |
CN103488664A (en) * | 2013-05-03 | 2014-01-01 | 中国传媒大学 | Image retrieval method |
CN104679771A (en) * | 2013-11-29 | 2015-06-03 | 阿里巴巴集团控股有限公司 | Individual data searching method and device |
CN104462357A (en) * | 2014-12-08 | 2015-03-25 | 百度在线网络技术(北京)有限公司 | Method and device for realizing personalized search |
CN104484380A (en) * | 2014-12-09 | 2015-04-01 | 百度在线网络技术(北京)有限公司 | Personalized search method and personalized search device |
CN104598611A (en) * | 2015-01-29 | 2015-05-06 | 百度在线网络技术(北京)有限公司 | Method and system for sequencing search entries |
Also Published As
Publication number | Publication date |
---|---|
CN106095983A (en) | 2016-11-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106095983B (en) | A kind of similarity based on personalized deep neural network determines method and device | |
AU2017268661B2 (en) | Large-scale image search and tagging using image-to-topic embedding | |
AU2017268662B2 (en) | Dense image tagging via two-stage soft topic embedding | |
US9147154B2 (en) | Classifying resources using a deep network | |
CN104123332B (en) | The display methods and device of search result | |
CN104598611B (en) | The method and system being ranked up to search entry | |
CN102955807B (en) | A kind of search method and device of related information | |
US20110055238A1 (en) | Methods and systems for generating non-overlapping facets for a query | |
CN104915399A (en) | Recommended data processing method based on news headline and recommended data processing method system based on news headline | |
CN106599194A (en) | Label determining method and device | |
CN112131261B (en) | Community query method and device based on community network and computer equipment | |
Na et al. | User identification based on multiple attribute decision making in social networks | |
CN116521906B (en) | Meta description generation method, device, equipment and medium thereof | |
CN107330010B (en) | Background path blasting method based on machine learning | |
CN111488524A (en) | Attention-oriented semantic-sensitive label recommendation method | |
CN113806630A (en) | Attention-based multi-view feature fusion cross-domain recommendation method and device | |
CN107506426A (en) | A kind of implementation method of intelligent television automated intelligent response robot | |
CN107577786A (en) | A kind of matrix decomposition recommendation method based on joint cluster | |
Li et al. | Coherent and controllable outfit generation | |
Suresh et al. | Sentiment classification using decision tree based feature selection | |
Kedia et al. | Keep learning: Self-supervised meta-learning for learning from inference | |
CN113312479B (en) | Cross-domain false news detection method | |
Bouhini et al. | Integrating user's profile in the query model for Social Information Retrieval | |
JP6495206B2 (en) | Document concept base generation device, document concept search device, method, and program | |
CN110262906B (en) | Interface label recommendation method and device, storage medium and electronic equipment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |