CN110162191A - A kind of expression recommended method, device and storage medium - Google Patents
A kind of expression recommended method, device and storage medium Download PDFInfo
- Publication number
- CN110162191A CN110162191A CN201910267896.2A CN201910267896A CN110162191A CN 110162191 A CN110162191 A CN 110162191A CN 201910267896 A CN201910267896 A CN 201910267896A CN 110162191 A CN110162191 A CN 110162191A
- Authority
- CN
- China
- Prior art keywords
- expression
- recommended
- target
- label
- similarity
- 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
- 238000000034 method Methods 0.000 title claims abstract description 50
- 238000003860 storage Methods 0.000 title claims abstract description 14
- 239000013598 vector Substances 0.000 claims description 70
- 238000012549 training Methods 0.000 claims description 59
- 238000003062 neural network model Methods 0.000 claims description 38
- 230000015654 memory Effects 0.000 claims description 28
- 230000011218 segmentation Effects 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 5
- 238000010586 diagram Methods 0.000 description 21
- 101150054987 ChAT gene Proteins 0.000 description 14
- 101100203187 Mus musculus Sh2d3c gene Proteins 0.000 description 14
- 238000004364 calculation method Methods 0.000 description 13
- 238000013528 artificial neural network Methods 0.000 description 7
- 230000000694 effects Effects 0.000 description 7
- 230000003068 static effect Effects 0.000 description 6
- 238000004891 communication Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 5
- 230000003993 interaction Effects 0.000 description 4
- 206010028916 Neologism Diseases 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 3
- 230000002452 interceptive effect Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 230000008451 emotion Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000013515 script Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 108010074506 Transfer Factor Proteins 0.000 description 1
- 238000000429 assembly Methods 0.000 description 1
- 230000000712 assembly Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000000151 deposition Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000009545 invasion Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000036651 mood Effects 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 239000000725 suspension Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F3/00—Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
- G06F3/01—Input arrangements or combined input and output arrangements for interaction between user and computer
- G06F3/02—Input arrangements using manually operated switches, e.g. using keyboards or dials
- G06F3/023—Arrangements for converting discrete items of information into a coded form, e.g. arrangements for interpreting keyboard generated codes as alphanumeric codes, operand codes or instruction codes
- G06F3/0233—Character input methods
- G06F3/0237—Character input methods using prediction or retrieval techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/01—Social networking
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Human Computer Interaction (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the invention provides a kind of expression recommended method, device and storage mediums；Wherein, method comprises determining that the target text based on client input；The corresponding expression label of expression element to be recommended is obtained, the similarity of the target text Yu the expression label is obtained；Based on the similarity, target expression element being chosen from the expression element to be recommended and carries out expression recommendation, the target expression element includes: the corresponding expression element of non-fully identical with target text expression label.
Description
Technical field
The present invention relates to the communication technology more particularly to a kind of expression recommended methods, device and storage medium.
Background technique
With social and network continuous development, is linked up based on social client by expression and formed a kind of prevalence
Culture obtains good communication experience to enable to interact both sides, such social client also supports expression recommendation function, i.e.,
Recommend suitable expression during user interaction for user.
Recognition with Recurrent Neural Network (RNN, Recurrent Neural Network) and convolutional Neural net are relied in the related technology
The learning model of network (CNN, Convolutional Neural Networks) carries out expression recommendation, this needs a large amount of data
Training, and user's expression is not consistent based on different interactive scene demands, expression recommends accuracy low.
Summary of the invention
The embodiment of the present invention provides a kind of expression recommended method, device and storage medium, can be realized accurately pushing away for expression
It recommends.
The technical solution of the embodiment of the present invention is achieved in that
On the one hand, the embodiment of the present invention provides a kind of expression recommended method, comprising:
Determine the target text inputted based on client；
Obtain the corresponding expression label of expression element to be recommended；
Obtain the similarity of the target text Yu the expression label；
Based on the similarity, target expression element is chosen from the expression element to be recommended and carries out expression recommendation, institute
Stating target expression element includes: the corresponding expression element of non-fully identical with target text expression label.
On the one hand, the embodiment of the present invention provides a kind of expression recommendation apparatus, and described device includes:
Determination unit, for determining the target text inputted based on client；
Acquiring unit, for obtaining the corresponding expression label of expression element to be recommended；
Computing unit, for obtaining the similarity of the target text Yu the expression label；
Recommendation unit, for be based on the similarity, from the expression element to be recommended choose target expression element into
Row expression is recommended, and the target expression element includes: the corresponding expression of non-fully identical with target text expression label
Element.
On the one hand, the embodiment of the present invention provides a kind of expression recommendation apparatus, comprising:
Memory, for storing executable instruction；
Processor when for executing the executable instruction stored in the memory, is realized provided in an embodiment of the present invention
Expression recommended method.
The embodiment of the present invention provides a kind of storage medium, is stored with executable instruction, real when for causing processor to execute
Existing expression recommended method provided in an embodiment of the present invention.
The embodiment of the present invention has the advantages that
1) target text that the selection of target expression element is inputted based on user, so that the expression recommend chosen
The element interactive scene current with user is adapted, and improves the success rate that expression is recommended；
2) by calculating the target text of user's input and the similarity of each expression label, the target based on similarity is realized
The selection of expression element includes that expression label non-fully identical with target text is corresponding in selected target expression element
Expression element expands the optional range that can be used for the expression element recommended, and provide the user bigger expression selection space,
Greatly improve user experience.
Detailed description of the invention
Fig. 1 is the interface schematic diagram provided in an embodiment of the present invention for carrying out expression recommendation；
Fig. 2 is an optional configuration diagram of expression recommender system 100 provided in an embodiment of the present invention；
Fig. 3 is the composed structure schematic diagram of expression recommendation apparatus provided in an embodiment of the present invention；
Fig. 4 is the flow diagram of expression recommended method provided in an embodiment of the present invention；
Fig. 5 is the schematic illustration of neural network model provided in an embodiment of the present invention；
Fig. 6 is the structural schematic diagram of neural network model provided in an embodiment of the present invention；
Fig. 7 is the structural schematic diagram of neural network model provided in an embodiment of the present invention；
Fig. 8 is the interface schematic diagram provided in an embodiment of the present invention for carrying out expression recommendation；
Fig. 9 is the flow diagram of expression recommended method provided in an embodiment of the present invention；
Figure 10 is the flow diagram of expression recommended method provided in an embodiment of the present invention；
Figure 11 is the schematic diagram of neural network model provided in an embodiment of the present invention training；
Figure 12 is the composed structure schematic diagram of expression recommendation apparatus 120 provided in an embodiment of the present invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing to the present invention make into
It is described in detail to one step, described embodiment is not construed as limitation of the present invention, and those of ordinary skill in the art are not having
All other embodiment obtained under the premise of creative work is made, shall fall within the protection scope of the present invention.
In the following description, it is related to " some embodiments ", which depict the subsets of all possible embodiments, but can
To understand, " some embodiments " can be the same subsets or different subsets of all possible embodiments, and can not conflict
In the case where be combined with each other.
In the following description, related term " first second " be only be the similar object of difference, do not represent needle
To the particular sorted of object, it is possible to understand that specific sequence or successively can be interchanged in ground, " first second " in the case where permission
Order, so that the embodiment of the present invention described herein can be implemented with the sequence other than illustrating or describing herein.
Unless otherwise defined, all technical and scientific terms used herein and belong to technical field of the invention
The normally understood meaning of technical staff is identical.Term used herein is intended merely to the purpose of the description embodiment of the present invention,
It is not intended to limit the present invention.
Before the embodiment of the present invention is further elaborated, to noun involved in the embodiment of the present invention and term
It is illustrated, noun involved in the embodiment of the present invention and term are suitable for following explanation.
1) expression element, i.e. expression, after social application is active, a kind of pop culture of formation is specific to express
Emotion, such as the emotion that is showed in expression user's face or posture；In practical applications, expression can be divided into symbol expression,
Static images expression, dynamic picture expression etc., for example, expression can to express the faces of the various moods of user as material, or
With popular at present star, animation, video display screenshot etc. for material, then mix a series of texts etc. to match.
2) expression label, the corresponding text description of expression element, generally a word；In practical applications, shown
Corresponding expression can carry expression label, and the semantic content of expression label is co-expressed in conjunction with expression and expression label, also can be only
The semantic content of the expression label is expressed by expression.
3) it segments, also known as word cutting, refers to a character (including Chinese character, letter, number etc.) sequence being cut into one
Each and every one individual word.
4) stop words, to improve the word for recommending expression to generate effect that expression recommends efficiency to be filtered；Usually not
With meaning (only putting it into a complete sentence just has certain effect), for example, pronoun, article, adverbial word, Jie
Word and conjunction etc..
5) in response to the condition or state relied on for indicating performed operation, when the relied on condition of satisfaction
Or when state, performed one or more operations be can be in real time, it is possible to have the delay of setting；Do not saying especially
In the case where bright, there is no the limitations for executing sequencing for performed multiple operations.
In some embodiments, it when user is interacted based on social client (such as wechat, QQ), is such as chatted by wechat
It, user is based on client and inputs chat content, and the word that client inputs user is corresponding with each expression in expression packet
Label is matched, then expression corresponding to the identical label of word for determining and inputting recommends determining expression
To user.Fig. 1 is the interface schematic diagram provided in an embodiment of the present invention for carrying out expression recommendation, referring to Fig. 1, when user is based on currently
Chat client input " happy " when, client obtains corresponding expression by word match and is recommended.However, user is past
Toward that can not remember expression label corresponding to each expression, such as when user thinks expression thanks, input " sense in interactive process
Thank " when, expression label can not be matched to as the expression of " thanks " by using the above-mentioned expression way of recommendation just, and expression recommends limitation very
Greatly, more expression selections can not be provided the user with, user experience is low.
In some embodiments, when user is interacted based on client, user inputs text, client in the client
The target text for determining user's input obtains the corresponding expression label of expression element to be recommended, expression label and expression to be recommended
There are one-to-one relationships for element；Obtain the similarity of target text and each expression label；Based on similarity, from expression to be recommended
Target expression element is chosen in element and carries out expression recommendation, and target expression element includes at least: non-fully identical as target text
The corresponding expression element of expression label.In this way, expanding the optional range that can be used for the expression element recommended, provide the user with
Bigger expression selects space, greatly improves user experience.
Next expression recommender system provided in an embodiment of the present invention, device and method are illustrated respectively.
Fig. 2 is an optional configuration diagram of expression recommender system 100 provided in an embodiment of the present invention, referring to fig. 2,
An exemplary application is supported to realize, is provided in terminal (including terminal 400-1 and terminal 400-2) and is interacted for user social contact
Client, terminal by network 300 connect server 200, network 300 can be wide area network or local area network, or be
Combination realizes that data are transmitted using Radio Link.
Terminal (terminal 400-1 and/or terminal 400-2) for the determining target text inputted based on client, and is obtained
The corresponding expression label of expression element to be recommended is taken, the expression label and the expression element to be recommended, which exist to correspond, to close
System；
Server 200 for calculating the similarity of the target text Yu each expression label, and returns to similarity meter
Result is calculated to terminal；
The terminal is also used to based on similarity calculation as a result, choosing target expression from the expression element to be recommended
Element carries out expression recommendation, and the target expression element includes at least: expression label non-fully identical with the target text
Corresponding expression element.
In some embodiments, the client being arranged in terminal can be by the target text of input, and the table to be recommended of acquisition
The corresponding expression label of feelings element is sent to server, and the expression label and the expression element to be recommended, which exist, to be corresponded
Relationship；Server calculates separately the similarity of the target text Yu each expression label, and returns to similarity calculation result
To client；Client is based on similarity calculation and carries out as a result, choosing target expression element from the expression element to be recommended
Expression is recommended, and the target expression element includes at least: the corresponding table of non-fully identical with target text expression label
Feelings element.
Next expression recommendation apparatus provided in an embodiment of the present invention is illustrated.The expression of the embodiment of the present invention is recommended
Device can be implemented in a variety of manners, such as: it is individually real by smart phone, tablet computer and desktop computer terminal or server
It applies, or by terminal, server coordinated implementation.Expression recommendation apparatus provided in an embodiment of the present invention may be embodied as hardware or
The mode of software and hardware combining illustrates the various exemplary implementations of device provided in an embodiment of the present invention below.
It elaborates below to the hardware configuration of the expression recommendation apparatus of the embodiment of the present invention, Fig. 3 is that the present invention is implemented
The composed structure schematic diagram for the expression recommendation apparatus that example provides, it will be understood that Fig. 3 illustrate only the example of expression recommendation apparatus
Property structure rather than entire infrastructure, can be implemented the part-structure or entire infrastructure shown in Fig. 3 as needed.
Expression recommendation apparatus 20 provided in an embodiment of the present invention includes: at least one processor 201, memory 202, user
Interface 203 and at least one network interface 204.Various components in expression recommendation apparatus 20 are coupled in by bus system 205
Together.It is appreciated that bus system 205 is for realizing the connection communication between these components.It includes data that bus system 205, which is removed,
It further include power bus, control bus and status signal bus in addition except bus.It, will in Fig. 3 but for the sake of clear explanation
Various buses are all designated as bus system 205.
Wherein, user interface 203 may include display, keyboard, mouse, trace ball, click wheel, key, button, sense of touch
Plate or touch screen etc..
It is appreciated that memory 202 can be volatile memory or nonvolatile memory, may also comprise volatibility and
Both nonvolatile memories.Wherein, nonvolatile memory can be read-only memory (ROM, Read Only Memory),
Programmable read only memory (PROM, Programmable Read-Only Memory), Erasable Programmable Read Only Memory EPROM
(EPROM, Erasable Programmable Read-Only Memory), flash memory (Flash Memory) etc..Volatibility is deposited
Reservoir can be random access memory (RAM, Random Access Memory), be used as External Cache.By showing
Example property but be not restricted explanation, the RAM of many forms is available, such as static random access memory (SRAM, Static
Random Access Memory), synchronous static random access memory (SSRAM, Synchronous Static Random
Access Memory).The memory 202 of description of the embodiment of the present invention is intended to include depositing for these and any other suitable type
Reservoir.
Memory 202 in the embodiment of the present invention can storing data to support the operation of terminal (such as 400-1).These numbers
According to example include: any computer program for being operated on terminal (such as 400-1), such as operating system and application program.
Wherein, operating system includes various system programs, such as ccf layer, core library layer, driving layer etc., for realizing various basic industry
It is engaged in and handles hardware based task.Application program may include various application programs.
As the example that method provided in an embodiment of the present invention uses software and hardware combining to implement, the embodiment of the present invention is provided
Method can be embodied directly in and combined by the software module that processor 201 executes, software module can be located in storage medium,
Storage medium is located at memory 202, and processor 201 reads the executable instruction that software module includes in memory 202, in conjunction with must
The hardware (e.g., including processor 201 and the other assemblies for being connected to bus 205) wanted is completed provided in an embodiment of the present invention
Expression recommended method.
As an example, processor 201 can be a kind of IC chip, and the processing capacity with signal, for example, it is general
Processor, digital signal processor (DSP, Digital Signal Processor) or other programmable logic device are divided
Vertical door or transistor logic, discrete hardware components etc., wherein general processor can be microprocessor or any normal
The processor etc. of rule.
As the example that expression recommended method provided in an embodiment of the present invention uses hardware to implement, the embodiment of the present invention is mentioned
The processor 201 of hardware decoding processor form can be directly used to execute completion, for example, one or more in the method for confession
Application specific integrated circuit (ASIC, Application Specific Integrate d Circuit), DSP, it may be programmed and patrol
Collect device (PLD, Programmable Logic Device), Complex Programmable Logic Devices (CPLD, Complex
Programmable Logic Device), field programmable gate array (FPGA, Field-Programmable Gate
Array) or other electronic components execute and realize method provided in an embodiment of the present invention.
Memory 202 in the embodiment of the present invention is for storing various types of data to support expression recommendation apparatus 20
Operation.The example of these data includes: any executable instruction for operating on expression recommendation apparatus 20, such as executable to refer to
It enables, realizes that the program of the expression recommended method of the embodiment of the present invention may be embodied in executable instruction.
Next expression recommended method provided in an embodiment of the present invention is illustrated.Fig. 4 is that the embodiment of the present invention provides
Expression recommended method flow diagram, in some embodiments, which can be implemented by terminal, such as pass through figure
Terminal 400-1 in 2 is implemented, and referring to fig. 4, expression recommended method provided in an embodiment of the present invention includes:
Step 301: terminal determines the target text inputted based on client.
In practical applications, the client of social application is provided in terminal, as instant communication client (wechat/QQ),
At least one expression element has been locally stored in microblogging client etc., client, and there are corresponding one or more for each expression element
A expression label, user can be interacted by client, pass through the text information of client input interaction.
In some embodiments, terminal can be determined as follows the target text based on client input, comprising: eventually
End obtains participle dictionary, which includes at least the corresponding expression label of expression element to be recommended；It is right based on participle dictionary
Text based on client input carries out word segmentation processing, obtains corresponding word sequence；Choose the last one word conduct in word sequence
Target text.
In actual implementation, the participle dictionary for being segmented to the text of input may also include preset participle word
Library such as includes Jieba dictionary, in conjunction with preset dictionary and the corresponding expression label of expression element to be recommended of segmenting to the text of input
This is segmented, and the term habit that can be bonded user preferably segments, and raising segments some neologisms and network word
Identification, and then more accurately carry out expression recommendation.Such as: the text that user is inputted based on client is that " 666, today is really too
Open gloomy ", in the input text " 666 " and " opening gloomy " be cyberspeak, according to participle dictionary in do not include expression element
Corresponding expression label, then the word segmentation result obtained be " 6 | 6 | 6 |, | today | it is genuine | too | open | it is gloomy ", be based on the word segmentation result
The last one word of obtained user's input, terminal obviously can not match and recommend suitable expression, using the embodiment of the present invention
The above-mentioned participle dictionary provided carry out obtained result after text participle be " 666 |, | today | it is genuine | too | open gloomy ", it is clear that point
Word is more accurate.
In some embodiments, terminal further includes construction point before being segmented using participle dictionary to the text of input
The operation of word dictionary, in practical applications, terminal obtain the expression element of terminal local storage, determine the expression element institute of storage
Default participle dictionary (such as Jieba dictionary) is added in determining expression label by corresponding expression label, is formed above-mentioned for defeated
The participle dictionary that the text entered is segmented.
In some embodiments, terminal, can also base after the text inputted based on participle dictionary to user is segmented
The filtering of stop words is carried out in the participle dictionary constructed, is further realized to the target for carrying out the selection of target expression element
The screening of text；For example, for word segmentation result " 666 |, | today | it is genuine | too | open gloomy | " it is last before filtering stop words
One word is " ", and filtering the last one word after stop words is " opening gloomy ", in this way, improving the accurate of determining target text
Degree.
Step 302: terminal obtains the corresponding expression label of expression element to be recommended.
In some embodiments, multiple expression elements are stored in terminal, multiple expression elements of storage are as to be recommended
Expression element, each expression element to be recommended correspond at least one expression label, in some embodiments, expression label with wait push away
Expression element is recommended there are one-to-one relationship, if the corresponding expression label of certain expression element to be recommended is " happy ", at other
In embodiment, an expression element to be recommended corresponds to two or more expression labels, such as certain expression element pair to be recommended
The expression label answered is " thanks " and " thanks ".
Step 303: the similarity of terminal calculating target text and expression label.
In some embodiments, the neural network model that terminal can be obtained by training calculates target text and each expression mark
The similarity of label, specifically, can calculate the similarity of target text Yu each expression label in the following way: terminal is literary by target
This input neural network model obtains corresponding semantic vector, obtains the corresponding label vector of each expression label；It calculates separately
The included angle cosine value of the semantic vector and each label vector that arrive, obtains the similarity of target text Yu each expression label；Wherein, refreshing
Text is inputted based on the history of target user through network model and the training of preset first training sample set obtains.
In actual implementation, terminal can train above-mentioned neural network model in the following way: initialization neural network mould
Input layer, middle layer and the output layer that type includes；The second training sample set is constructed, the second training sample set includes: history
The first sample word and corresponding first semantic vector in text are inputted, by preset second sample word and corresponding second semanteme
The first training sample set that vector is constituted；Terminal is respectively using first sample word and the second sample word as input, with corresponding
As output, training neural network model is corresponding semantic according to the text output of input for first semantic vector and the second semantic vector
The performance of vector.
Here, the training for the performance for corresponding to semantic vector according to the text output of input to neural network model is said
It is bright, it in some embodiments, can implement in the following way: construct the loss function of corresponding neural network model, be based on institute's structure
The loss function built determines the error signal of neural network model, and the error signal is reversely passed in neural network model
It broadcasts, and updates each layer of model parameter during propagation.In practical applications, loss function can be neural network mould
The conditional probability of text corresponding to the semantic vector of type output.
Here backpropagation is illustrated, training sample data is input to the input layer of neural network model, passed through
Hidden layer finally reaches output layer and exports as a result, this is the propagated forward process of neural network model, due to neural network mould
The output result of type and actual result have error, then calculate the error between output result and actual value, and by the error from defeated
Layer is to hidden layer backpropagation out, until input layer is traveled to, during backpropagation, according to error transfer factor model parameter
Value；The continuous iteration above process, until convergence.
First sample word in history input text is that the history of target user is inputted text to segment to obtain, due to user
Between the text information that the interacts more life-stylize compared with general article, have the expression and some native dialects of many common sayings
Expression can be improved the semanteme of term vector in this way, the history input text of user to be used as to the data of neural network model training
The degree of association is more bonded interaction (chat) scene of user；In practical applications, it is understood that there may be the history of user input text compared with
It is few, so that the case where model training data deficiencies therefore, trained word (the second sample word) and corresponding term vector can be preset
The training sample of (the second semantic vector) as neural network model in the embodiment of the present invention, avoids model training data deficiencies
Situation occurs, and preset second sample word and corresponding second semantic vector can be based on " encyclopaedia " in network, " news articles "
Etc. obtaining.
Fig. 5 is that the schematic illustration of neural network model provided in an embodiment of the present invention passes through neural network referring to Fig. 5
Model realization, specifically can be by inputting a word x according to given input text prediction contextk, output window size C
In each word probability.For example, for sentence " I drive my car to the store ", if " car " is sample word,
I.e. as training input data, correspondingly, group of words { " I ", " drive ", " my ", " to ", " the ", " store " } is then defeated
Out.
Fig. 6 is the structural schematic diagram of neural network model provided in an embodiment of the present invention, referring to Fig. 6, neural network model
Including input layer, hidden layer and output layer, in actual implementation, due to character string can not directly as the input and output of model,
Therefore, it needs to carry out one-hot encoding coding to input text before text (word) is inputted neural network model, obtains corresponding input
The one-hot encoding of text will also input word and be expressed as an one-hot vector, and referring to Fig. 6, the dimension of vector is for training
The word amount of the dictionary of neural network model, for example, if including 1000 words in training dictionary, then word is encoded to 1000
Dimension, the value of the corresponding position of word is 1, and other positions 0, in practical applications, hidden layer carry out feature to input vector
It extracts, the line number of hidden layer is 1000, and columns is characterized number.
Fig. 7 is the structural schematic diagram of neural network model provided in an embodiment of the present invention, is represented referring to Fig. 7, input vector x
The one-hot encoding of some word, V are the dimension of corresponding input vector, corresponding output vector { y1…yC, input layer and hidden layer it
Between the i-th row of weight matrix W represent the weight of i-th of word in vocabulary, the output that each output word vector has N × V to tie up
Vector W ', the shared weight of each word in output layer, output layer corresponding for softmax function generate the C word
Multinomial distribution, i.e., the value be the C export word j-th of node the following formula of probability size:
Wherein, wherein Wc,jIndicate j-th of word of c-th of panel of output layer；WO,cFor c-th of list for exporting context
Word；WIIt is unique input word；yc,jIt is the output valve (probability) of j-th of neuron node on c-th of panel of output layer；uc,j
Defeated is the output valve of layer j-th of node on c-th of panel out, and all panels of output layer share same weight matrix；Wherein,
Panel is the combination of the neuron of each context words of expression of output layer.
It in some embodiments, can be using the side of " negative sampling " in order to improve the training speed to neural network model
Method updates all weights different from each training sample originally, and it is small that negative sampling allows a training sample only to update one every time
Partial weight, when the word of input is twin target word-context, label is set as 1, and (context here is also one
Word), in addition arbitrarily take k to the non-context of target word-as negative sample, label is set as 0, so can reduce gradient and declined
Calculation amount in journey.
It should be noted that in some embodiments, the calculating of target text and expression label similarity also can be by servicing
Device executes.
Step 304: terminal is based on similarity calculation and carries out as a result, choosing target expression element from expression element to be recommended
Expression is recommended, and target expression element includes: the corresponding expression element of non-fully identical with target text expression label.
Here, expression element included by target expression element is illustrated, in one embodiment, target expression element
Including at least the corresponding expression element of expression label non-fully identical with target text, that is, including at least and target text
Expression label of the similarity less than 1 corresponding to expression element, in practical applications, in target expression element, with target text
Expression element corresponding to this expression label of the similarity less than 1 can have multiple, and bigger table is so provided for user
Feelings select space, and certainly, in another embodiment, target expression element may also include and the identical expression mark of target text
Sign corresponding expression element, that is, include with the similarity of target text equal to 1 expression label corresponding to expression element,
In this way, providing the highest expression of compatible degree with input text for user, user experience is improved.
In some embodiments, terminal can realize the selection and recommendation of target expression element in the following way:
It, will based on similarity calculation as a result, the determining similarity with target text reaches the expression label of similarity threshold
The corresponding expression element to be recommended of determining expression label is as target expression element；Using the first display mode, pass through client
Target expression element is presented in end.
Here, the size of similarity threshold can be set based on actual needs, in practical applications, with target text
The expression label that similarity reaches similarity threshold can have it is multiple, for example, target text be " happiness ", the phase with target text
It include the following: " happy " like the expression label that degree reaches similarity threshold, corresponding similarity is 0.7313591241836548；
" excitement ", corresponding similarity are 0.5977063179016113；" gratified ", corresponding similarity are
0.5703485012054443；" excitement ", corresponding similarity are 0.5697213411331177；" heartily ", corresponding similar
Degree is 0.5673317909240723；" happiness ", corresponding similarity are 0.5251850485801697；When preset similarity
Threshold value be 0.57 when, choose expression label be " happy ", " excitement ", " gratified " corresponding expression element conduct to be recommended
Target expression element.In practical applications, the first display mode corresponds to the display effect of target expression element, the display that such as suspends,
Mobile display, static status display, default transparency show that Fig. 8 is the interface provided in an embodiment of the present invention for carrying out expression recommendation
Schematic diagram, referring to Fig. 8, " happiness " that terminal is inputted based on user is not present and " happiness " complete phase after carrying out similarity calculation
With expression label, take similarity to be greater than the corresponding expression element of expression label of threshold value, the expression of recommendation be respectively " happy ",
" excitement ", " gratified ".
In some embodiments, terminal can also realize the selection and recommendation of target expression element in the following way:
Recommend expression element to be ranked up as a result, treating according to the height of similarity based on similarity calculation, obtains wait push away
Recommend expression element sequence；Since expression element sequence to be recommended first expression element to be recommended, preset quantity is chosen
Expression element to be recommended is as target expression element；Using the second display mode, target expression element is presented by client.?
That is, do not go to consider the value of specific similarity during choosing target expression element, regardless of whether in the presence of with target text
Similarity be greater than threshold value expression label, selection preset quantity expression element to be recommended recommended, in this way, give user
Bigger expression is provided and selects space, guidance user carries out expression use, and user experience is good.
In some embodiments, the expression recommended method of the embodiment of the present invention can also be by terminal, server coordinated implementation, such as
Implemented by terminal 400-1 in Fig. 2 and server 200, the client for instant messaging is provided in terminal, Fig. 9 is this
The flow diagram for the expression recommended method that inventive embodiments provide, referring to Fig. 9, expression recommendation side provided in an embodiment of the present invention
Method includes:
Step 401: client receives the text of target user's input.
Here, target user is based on client and carries out chat interaction, the Text Entry input text shown based on client
This (i.e. chat content), for example, target user's input " my good excitement ".
Step 402: the last one word is as target word in client selection text.
In actual implementation, client realizes the selection of target word in the following way:
Client obtains the participle dictionary comprising multiple expression labels and default dictionary, and the participle dictionary based on acquisition is to defeated
Enter and segmented from text, and filter stop words, obtain word sequence, choosing the last one word in word sequence is target word.Here
Expression label corresponding to the expression element that expression label in the participle dictionary is locally stored for client.
Step 403: client obtains expression label corresponding to the expression element being locally stored.
In some embodiments, expression element and expression label are one-to-one relationship, and each expression element is one corresponding
Expression label.
Step 404: client sends similarity calculation and requests to server.
In actual implementation, the expression label of target word and acquisition is carried in similarity calculation request, for requesting service
The similarity of device calculating target word and each expression label.
Step 405: server obtains the term vector of corresponding target word and the label vector of corresponding expression label respectively.
In actual implementation, server by the obtained neural network model of target word input training, export corresponding word to
Amount；Neural network model inputs text based on the history of target user and the training of preset training sample set obtains, the nerve
The training process of network model is not repeated herein referring to foregoing description.In some embodiments, expression is stored in server
Corresponding label vector can be obtained by indexing expression label in the mapping table of label and label vector.
Step 406: server calculates the similarity of target word and each expression label, returns to similarity result to client.
Here, in actual implementation, server is obtained by calculating separately the included angle cosine value of term vector Yu each label vector
To the similarity of target word and each expression label.
Step 407: the determining similarity with target word of client reaches the expression label of similarity threshold.
In practical applications, similarity threshold can be set according to actual needs.
Step 408: expression element corresponding to determining expression label is presented using suspension display mode for client.
Figure 10 is that the flow diagram of expression recommended method provided in an embodiment of the present invention is passed through referring to Figure 10 with user
It is application scenarios that instant communication client, which carries out chat, after user inputs chat text, by carrying out text participle, is obtained most
A close word, and the corresponding term vector of the word is obtained, the cosine similarity of the term vector and expression label is calculated, by similarity
It sorts from high to low, chooses the expression label for being greater than " threshold value ", obtain the corresponding expression of expression label of selection and recommend use
Family, to realize that the expression to user is recommended.
Based on Figure 10, wherein step S1, the i.e. acquisition of term vector, the neural network model that can be specifically obtained by training
It obtains, and the training of the neural network model can be found in Figure 11, Figure 11 is the signal of model training provided in an embodiment of the present invention
Figure, the training sample of model include user history chat text information and in advance trained term vector, next based on figure
11 are respectively illustrated each operation.
In step s 2, it because Chinese is different from the particularity of English, before constructing term vector, needs to carry out text
Participle, a good participle effect can largely improve the effect of term vector training.So in addition to using default " Jieba
Other than dictionary, we also as a word, local expression corresponding label are loaded into participle tool participle ", can be improved pair
The identification of some neologisms and network word participle.Such as: " 666, today, genuine Tai Kaisen ", was not loaded with point of expression label to sentence
Word effect are as follows: " 6 | 6 | 6 |, | today | genuine | too | open | it is gloomy | ", the participle effect after loading expression text: " 666 |, | today
| it is genuine | too | open gloomy | ".
In step s3, term vector is initialized.Lack training data when user's chat message is few in order to prevent,
Some trained models can be preloaded into, the data of these model trainings are generally basede on online " encyclopaedia ", " news text
Chapter " etc., and because of the text information of chat more life-stylize compared with general article, there is the expression of many common sayings, additionally
Have some expression based on native dialect, so just need on existing term vector model, additional local chat message carry out into
The term vector training of one step is more bonded the chat scenario of user to improve the semantic association degree of term vector.
S4 is the training pattern of term vector, which is to predict context with current word, and neural network structure is referring to figure
6, for model training referring to foregoing description, be not described herein.
Using the above embodiment of the present invention on the basis of original Chinese word segmentation, expression label has been added as participle dictionary
A part, preferably text can be segmented, the word especially often occurred in some neologisms or expression has better participle
Performance；Compared to recommending based on the deep learning of user's mass data history chat message, and it is likely to result in invasion of privacy
Possibility, the training based on term vector can establish on existing extensive term vector collection, and complementarity is used in locally load
The training of family history chat message, and these supplement parts, can preferably embody the expression of some user individuals.
Continue to be illustrated the software implementation of expression recommendation apparatus provided in an embodiment of the present invention, Figure 12 is that the present invention is real
The composed structure schematic diagram for applying the expression recommendation apparatus 120 of example offer, referring to Figure 12, expression provided in an embodiment of the present invention is recommended
Device 120 includes:
Determination unit 121, for determining the target text inputted based on client；
Acquiring unit 122, for obtaining the corresponding expression label of expression element to be recommended；
Computing unit 123, for calculating the similarity of the target text Yu the expression label；
Recommendation unit 124, for being based on similarity calculation as a result, choosing target expression from the expression element to be recommended
Element carries out expression recommendation, and the target expression element includes at least: expression label non-fully identical with the target text
Corresponding expression element.
In some embodiments, the determination unit, is also used to obtain participle dictionary, and the participle dictionary includes at least institute
State the corresponding expression label of expression element to be recommended；
Based on the participle dictionary, word segmentation processing is carried out to the text inputted based on the client, obtains corresponding word
Sequence；
The word of the last one input in the word sequence is chosen as the target text.
In some embodiments, the recommendation unit is also used to based on similarity calculation as a result, determining literary with the target
This similarity reaches the expression label of similarity threshold, using the corresponding expression element to be recommended of determining expression label as institute
State target expression element；
Using the first display mode, the target expression element is presented by the client.
In some embodiments, the recommendation unit, be also used to based on similarity calculation as a result, according to similarity height
The expression element to be recommended is ranked up, expression element sequence to be recommended is obtained；
Since the expression element sequence to be recommended first expression element to be recommended, choose preset quantity wait push away
Expression element is recommended as the target expression element；
Using the second display mode, the target expression element is presented by the client.
In some embodiments, the computing unit is also used to the target text inputting neural network model, obtain
Corresponding semantic vector；The neural network model inputs text and preset first training sample based on the history of target user
Set training obtains；
Obtain the corresponding label vector of each expression label；
The included angle cosine value for calculating separately the semantic vector Yu each label vector obtains the target text and each
The similarity of the expression label.
In some embodiments, described device further includes training unit；
The training unit, for constructing the second training sample set, the second training sample set includes: described goes through
History inputs the first sample word and corresponding first semantic vector in text, by preset second sample word and corresponding second language
The first training sample set that adopted vector is constituted；
Respectively using the first sample word and the second sample word as input, with corresponding first semantic vector and the
Two semantic vectors correspond to the performance of semantic vector according to the text output of input as output, the training neural network model.
In some embodiments, the computing unit is also used to carry out one-hot encoding coding to the target text, obtain pair
Answer the one-hot encoding of the target text；
The one-hot encoding is inputted into the neural network model, export for characterize the target text it is semantic it is semantic to
Amount.
It need to be noted that: above is referred to the description of expression recommendation apparatus, be with above method description it is similar, together
The beneficial effect of method describes, and does not repeat them here, thin for undisclosed technology in expression recommendation apparatus described in the embodiment of the present invention
Section, please refers to the description of expression recommended method embodiment of the present invention.
The embodiment of the present invention also provides a kind of storage medium for being stored with executable instruction, wherein being stored with executable finger
It enables, when executable instruction is executed by processor, processor will be caused to execute expression recommended method provided in an embodiment of the present invention,
For example, expression recommended method as shown in Figure 4.
In some embodiments, storage medium can be FRAM, ROM, PROM, EPROM, EE PROM, flash memory, magnetic surface
The memories such as memory, CD or CD-ROM；Be also possible to include one of above-mentioned memory or any combination various equipment.
In some embodiments, executable instruction can use program, software, software module, the form of script or code,
By any form of programming language (including compiling or interpretative code, or declaratively or process programming language) write, and its
It can be disposed by arbitrary form, including be deployed as independent program or be deployed as module, component, subroutine or be suitble to
Calculate other units used in environment.
As an example, executable instruction can with but not necessarily correspond to the file in file system, can be stored in
A part of the file of other programs or data is saved, for example, being stored in hypertext markup language (H TML, Hyper Text
Markup Language) in one or more scripts in document, it is stored in the single file for being exclusively used in discussed program
In, alternatively, being stored in multiple coordinated files (for example, the file for storing one or more modules, subprogram or code section).
As an example, executable instruction can be deployed as executing in a calculating equipment, or it is being located at one place
Multiple calculating equipment on execute, or, be distributed in multiple places and by multiple calculating equipment of interconnection of telecommunication network
Upper execution.
The above, only the embodiment of the present invention, are not intended to limit the scope of the present invention.It is all in this hair
Made any modifications, equivalent replacements, and improvements etc. within bright spirit and scope, be all contained in protection scope of the present invention it
It is interior.
Claims (15)
1. a kind of expression recommended method, which is characterized in that the described method includes:
Determine the target text inputted based on client；
Obtain the corresponding expression label of expression element to be recommended；
Obtain the similarity of the target text Yu the expression label；
Based on the similarity, target expression element is chosen from the expression element to be recommended and carries out expression recommendation, the mesh
Mark expression element includes: the corresponding expression element of non-fully identical with target text expression label.
2. the method as described in claim 1, which is characterized in that the target text that the determination is inputted based on client, comprising:
Participle dictionary is obtained, the participle dictionary includes at least the corresponding expression label of the expression element to be recommended；
Based on the participle dictionary, word segmentation processing is carried out to the text inputted based on the client, obtains corresponding word sequence；
The last one word is chosen in the word sequence as the target text.
3. the method as described in claim 1, which is characterized in that it is described to be based on the similarity, from the expression member to be recommended
Target expression element is chosen in element carries out expression recommendation, comprising:
Based on the similarity, the determining similarity with the target text reaches the expression label of similarity threshold, will determine
The corresponding expression element to be recommended of expression label as the target expression element；
Using the first display mode, the target expression element is presented by the client.
4. the method as described in claim 1, which is characterized in that it is described to be based on the similarity, from the expression member to be recommended
Target expression element is chosen in element carries out expression recommendation, comprising:
Based on the similarity, the expression element to be recommended is ranked up according to the height of similarity, obtains table to be recommended
Feelings element sequence；
Since the expression element sequence to be recommended first expression element to be recommended, the table to be recommended of preset quantity is chosen
Feelings element is as the target expression element；
Using the second display mode, the target expression element is presented by the client.
5. the method as described in claim 1, which is characterized in that the phase for obtaining the target text and the expression label
Like degree, comprising:
The target text is inputted into neural network model, obtains corresponding semantic vector；The neural network model is based on mesh
The history input text of mark user and the training of preset first training sample set obtain；
Obtain the corresponding label vector of each expression label；
The included angle cosine value for obtaining the semantic vector Yu each label vector respectively, obtain the target text with it is each described
The similarity of expression label.
6. method as claimed in claim 5, which is characterized in that the method also includes:
The second training sample set is constructed, the second training sample set includes: the first sample in the history input text
This word and corresponding first semantic vector, described first be made of preset second sample word and corresponding second semantic vector
Training sample set；
Respectively using the first sample word and the second sample word as input, with corresponding first semantic vector and the second language
Adopted vector corresponds to the performance of semantic vector according to the text output of input as output, the training neural network model.
7. method as claimed in claim 5, which is characterized in that it is described that the target text is inputted into neural network model, it obtains
To corresponding semantic vector, comprising:
One-hot encoding coding is carried out to the target text, obtains the one-hot encoding for corresponding to the target text；
The one-hot encoding is inputted into the neural network model, exports the semantic vector for characterizing the target text semanteme.
8. a kind of expression recommendation apparatus, which is characterized in that described device includes:
Determination unit, for determining the target text inputted based on client；
Acquiring unit, for obtaining the corresponding expression label of expression element to be recommended；
Computing unit, for obtaining the similarity of the target text Yu the expression label；
Recommendation unit chooses target expression element from the expression element to be recommended and carries out table for being based on the similarity
Feelings are recommended, and the target expression element includes: the corresponding expression element of non-fully identical with target text expression label.
9. device as claimed in claim 8, which is characterized in that
The determination unit, for obtaining participle dictionary, it is corresponding that the participle dictionary includes at least the expression element to be recommended
Expression label；
Based on the participle dictionary, word segmentation processing is carried out to the text inputted based on the client, obtains corresponding word sequence；
The last one word is chosen in the word sequence as the target text.
10. device as claimed in claim 8, which is characterized in that
The recommendation unit, for being based on the similarity, the determining similarity with the target text reaches similarity threshold
Expression label, using the corresponding expression element to be recommended of determining expression label as the target expression element；
Using the first display mode, the target expression element is presented by the client.
11. device as claimed in claim 8, which is characterized in that
The recommendation unit carries out the expression element to be recommended according to the height of similarity for being based on the similarity
Sequence, obtains expression element sequence to be recommended；
Since the expression element sequence to be recommended first expression element to be recommended, the table to be recommended of preset quantity is chosen
Feelings element is as the target expression element；
Using the second display mode, the target expression element is presented by the client.
12. device as claimed in claim 8, which is characterized in that
The computing unit obtains corresponding semantic vector for the target text to be inputted neural network model；The mind
Text is inputted based on the history of target user through network model and the training of preset first training sample set obtains；
Obtain the corresponding label vector of each expression label；
The included angle cosine value for obtaining the semantic vector Yu each label vector respectively, obtain the target text with it is each described
The similarity of expression label.
13. device as claimed in claim 12, which is characterized in that described device further includes training unit；
The training unit, for constructing the second training sample set, the second training sample set includes: that the history is defeated
Enter the first sample word and corresponding first semantic vector in text, from preset second sample word and it is corresponding second it is semantic to
Measure the first training sample set constituted；
Respectively using the first sample word and the second sample word as input, with corresponding first semantic vector and the second language
Adopted vector corresponds to the performance of semantic vector according to the text output of input as output, the training neural network model.
14. a kind of expression recommendation apparatus, which is characterized in that described device includes:
Memory, for storing executable instruction；
Processor when for executing the executable instruction stored in the memory, is realized described in any one of claim 1 to 7
Expression recommended method.
15. a kind of storage medium, which is characterized in that the storage medium is stored with executable instruction, for causing processor to be held
When row, the described in any item expression recommended methods of claim 1 to 7 are realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910267896.2A CN110162191B (en) | 2019-04-03 | 2019-04-03 | Expression recommendation method, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910267896.2A CN110162191B (en) | 2019-04-03 | 2019-04-03 | Expression recommendation method, device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110162191A true CN110162191A (en) | 2019-08-23 |
| CN110162191B CN110162191B (en) | 2024-08-06 |
Family
ID=67638473
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910267896.2A Active CN110162191B (en) | 2019-04-03 | 2019-04-03 | Expression recommendation method, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110162191B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110619357A (en) * | 2019-08-29 | 2019-12-27 | 北京搜狗科技发展有限公司 | Picture processing method and device and electronic equipment |
| CN111460177A (en) * | 2020-03-27 | 2020-07-28 | 北京奇艺世纪科技有限公司 | Method and device for searching film and television expression, storage medium and computer equipment |
| CN111507104A (en) * | 2020-03-19 | 2020-08-07 | 北京百度网讯科技有限公司 | Method and device for establishing label labeling model, electronic equipment and readable storage medium |
| CN111931494A (en) * | 2020-08-10 | 2020-11-13 | 北京字节跳动网络技术有限公司 | Method, apparatus, electronic device, and medium for generating prediction information |
| CN112748974A (en) * | 2020-08-05 | 2021-05-04 | 腾讯科技（深圳）有限公司 | Information display method, device, equipment and storage medium based on session |
| CN112905079A (en) * | 2019-11-19 | 2021-06-04 | 北京搜狗科技发展有限公司 | Data processing method, device and medium |
| CN113204703A (en) * | 2021-05-10 | 2021-08-03 | 北京明略昭辉科技有限公司 | Material recommendation method, device, equipment and storage medium |
| CN113674008A (en) * | 2020-05-14 | 2021-11-19 | 北京达佳互联信息技术有限公司 | Directional label recommendation method, device, server and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20110026218A (en) * | 2009-09-07 | 2011-03-15 | 동국대학교 산학협력단 | Apparatus and method for inputting text message and its program stored in recording medium |
| CN104063427A (en) * | 2014-06-06 | 2014-09-24 | 北京搜狗科技发展有限公司 | Expression input method and device based on semantic understanding |
| CN106095983A (en) * | 2016-06-20 | 2016-11-09 | 北京百度网讯科技有限公司 | A kind of similarity based on personalized deep neural network determines method and device |
| CN106888158A (en) * | 2017-02-28 | 2017-06-23 | 努比亚技术有限公司 | A kind of instant communicating method and device |
| CN107423277A (en) * | 2016-02-16 | 2017-12-01 | 中兴通讯股份有限公司 | A kind of expression input method, device and terminal |
-
2019
- 2019-04-03 CN CN201910267896.2A patent/CN110162191B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20110026218A (en) * | 2009-09-07 | 2011-03-15 | 동국대학교 산학협력단 | Apparatus and method for inputting text message and its program stored in recording medium |
| CN104063427A (en) * | 2014-06-06 | 2014-09-24 | 北京搜狗科技发展有限公司 | Expression input method and device based on semantic understanding |
| CN107423277A (en) * | 2016-02-16 | 2017-12-01 | 中兴通讯股份有限公司 | A kind of expression input method, device and terminal |
| CN106095983A (en) * | 2016-06-20 | 2016-11-09 | 北京百度网讯科技有限公司 | A kind of similarity based on personalized deep neural network determines method and device |
| CN106888158A (en) * | 2017-02-28 | 2017-06-23 | 努比亚技术有限公司 | A kind of instant communicating method and device |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110619357A (en) * | 2019-08-29 | 2019-12-27 | 北京搜狗科技发展有限公司 | Picture processing method and device and electronic equipment |
| CN110619357B (en) * | 2019-08-29 | 2022-03-04 | 北京搜狗科技发展有限公司 | Picture processing method and device and electronic equipment |
| CN112905079A (en) * | 2019-11-19 | 2021-06-04 | 北京搜狗科技发展有限公司 | Data processing method, device and medium |
| CN112905079B (en) * | 2019-11-19 | 2022-12-13 | 北京搜狗科技发展有限公司 | Data processing method, device and medium |
| CN111507104A (en) * | 2020-03-19 | 2020-08-07 | 北京百度网讯科技有限公司 | Method and device for establishing label labeling model, electronic equipment and readable storage medium |
| US11531813B2 (en) | 2020-03-19 | 2022-12-20 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Method, electronic device and readable storage medium for creating a label marking model |
| CN111460177B (en) * | 2020-03-27 | 2023-12-15 | 北京奇艺世纪科技有限公司 | Video expression search method and device, storage medium and computer equipment |
| CN111460177A (en) * | 2020-03-27 | 2020-07-28 | 北京奇艺世纪科技有限公司 | Method and device for searching film and television expression, storage medium and computer equipment |
| CN113674008A (en) * | 2020-05-14 | 2021-11-19 | 北京达佳互联信息技术有限公司 | Directional label recommendation method, device, server and storage medium |
| CN112748974A (en) * | 2020-08-05 | 2021-05-04 | 腾讯科技（深圳）有限公司 | Information display method, device, equipment and storage medium based on session |
| CN112748974B (en) * | 2020-08-05 | 2024-04-16 | 腾讯科技（深圳）有限公司 | Information display method, device, equipment and storage medium based on session |
| CN111931494A (en) * | 2020-08-10 | 2020-11-13 | 北京字节跳动网络技术有限公司 | Method, apparatus, electronic device, and medium for generating prediction information |
| US11842163B2 (en) | 2020-08-10 | 2023-12-12 | Beijing Bytedance Network Technology Co., Ltd. | Method and apparatus for generating prediction information, and electronic device and medium |
| CN113204703A (en) * | 2021-05-10 | 2021-08-03 | 北京明略昭辉科技有限公司 | Material recommendation method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110162191B (en) | 2024-08-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110162191A (en) | A kind of expression recommended method, device and storage medium | |
| US9934515B1 (en) | Content recommendation system using a neural network language model | |
| US11487946B2 (en) | Content editing using content modeling and semantic relevancy scoring | |
| CN104781813B (en) | Grammar model for structured search queries | |
| CN112632385A (en) | Course recommendation method and device, computer equipment and medium | |
| CN111090756B (en) | Artificial intelligence-based multi-target recommendation model training method and device | |
| CN112948708B (en) | Short video recommendation method | |
| CN113011186B (en) | Named entity recognition method, named entity recognition device, named entity recognition equipment and computer readable storage medium | |
| US10223440B2 (en) | Question and answer system emulating people and clusters of blended people | |
| CN111414561B (en) | Method and device for presenting information | |
| CN116010684A (en) | Article recommendation method, device and storage medium | |
| US9129216B1 (en) | System, method and apparatus for computer aided association of relevant images with text | |
| CN111931054B (en) | Sequence recommendation method and system based on improved residual error structure | |
| CN110321918A (en) | The method of public opinion robot system sentiment analysis and image labeling based on microblogging | |
| CN109783824A (en) | Interpretation method, device and storage medium based on translation model | |
| US11653071B2 (en) | Responsive video content alteration | |
| CN114175018A (en) | New word classification technique | |
| CN114417058A (en) | Video material screening method and device, computer equipment and storage medium | |
| CN116955591A (en) | Recommendation language generation method, related device and medium for content recommendation | |
| CN113869068A (en) | Scene service recommendation method, device, equipment and storage medium | |
| CN117708428A (en) | Recommendation information prediction method and device and electronic equipment | |
| CN113360615B (en) | Dialogue recommendation method, system and equipment based on knowledge graph and time sequence characteristics | |
| WO2022246162A1 (en) | Content generation using target content derived modeling and unsupervised language modeling | |
| CN106293114B (en) | Predict the method and device of user's word to be entered | |
| Yang | Analysis of top box office film poster marketing scheme based on data mining and deep learning in the context of film marketing |
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 |