CN111737595A - Candidate word recommendation method, word stock ordering model training method and device - Google Patents
Candidate word recommendation method, word stock ordering model training method and device Download PDFInfo
- Publication number
- CN111737595A CN111737595A CN202010594333.7A CN202010594333A CN111737595A CN 111737595 A CN111737595 A CN 111737595A CN 202010594333 A CN202010594333 A CN 202010594333A CN 111737595 A CN111737595 A CN 111737595A
- Authority
- CN
- China
- Prior art keywords
- word
- search
- candidate
- party
- sorting
- 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
- 238000012549 training Methods 0.000 title claims abstract description 92
- 238000000034 method Methods 0.000 title claims abstract description 82
- 238000004590 computer program Methods 0.000 claims description 17
- 238000004821 distillation Methods 0.000 claims description 17
- 238000010801 machine learning Methods 0.000 claims description 16
- 238000001914 filtration Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 description 21
- 238000010586 diagram Methods 0.000 description 14
- 230000006870 function Effects 0.000 description 13
- 230000000694 effects Effects 0.000 description 12
- 238000003860 storage Methods 0.000 description 12
- 230000006872 improvement Effects 0.000 description 9
- 238000012545 processing Methods 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 6
- 238000012163 sequencing technique Methods 0.000 description 5
- 238000013459 approach Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 230000000750 progressive effect Effects 0.000 description 2
- 230000002035 prolonged effect Effects 0.000 description 2
- 208000001613 Gambling Diseases 0.000 description 1
- 241000545744 Hirudinea Species 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 230000003203 everyday effect Effects 0.000 description 1
- 230000009545 invasion Effects 0.000 description 1
- 230000005055 memory storage Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 229920001296 polysiloxane Polymers 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/237—Lexical tools
- G06F40/242—Dictionaries
-
- 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
- G06F16/9536—Search customisation based on social or collaborative filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/237—Lexical tools
- G06F40/247—Thesauruses; Synonyms
Abstract
The embodiment of the specification provides a candidate word recommendation method, a word stock ordering model training method and a device. The method comprises the following steps: receiving a search request sent by a user, wherein the search request comprises search words; selecting a first candidate word set to be recommended related to the search word from a first word bank according to the search word; selecting a second candidate word set to be recommended related to the search word from at least one second word bank according to the search word; sorting the first candidate word set to be recommended by adopting a first word stock sorting model to obtain a first sorting result; sorting the second candidate word set to be recommended by adopting a second word stock sorting model to obtain a second sorting result; performing merging and sorting according to the first sorting result and the second sorting result; and recommending the candidate words based on the combined and sorted result.
Description
Technical Field
The specification relates to the technical field of internet, in particular to a candidate word recommendation method, a word stock ordering model training method and a device.
Background
At present, with the popularization and development of computer technology and internet technology, in order to provide more convenient and faster information service for users, a search function is added in a page of an application program, keywords input by the users can be received through the search function, and a series of suggested words (namely, candidate words recommended to the users) are given, so that the users can search information according to the suggested words to meet the information search requirements of the users. Since some application programs incorporate the search service of the third party application in their search function, it is necessary to add candidate word search results of the third party application in the search suggestion words.
In the prior art, candidate word search results of third-party applications are mainly fused in the following way, the third-party application services are directly called, and candidate words returned by the third-party applications and candidate words applied by the third-party applications are fused and then displayed to a user, but the time consumed for searching links is prolonged in the way; in addition, a suggested word library of the third-party application can be independently established in a server of the self-application, and suggested word search logic of the third-party application can be trained, but the method can face the problems that a user log of the third-party application is lacked, the suggested word search logic of the third-party application cannot be trained by using a machine learning model, and the search ranking recommendation effect of the self-established third-party application search service is greatly reduced.
Based on the prior art, a candidate word recommendation scheme capable of reducing time consumption of searching links, minimizing search delay, and ensuring search sorting recommendation effect without depending on a user log of a third-party application is needed.
Disclosure of Invention
The embodiment of the specification provides a candidate word recommendation method, a word stock ordering model training method and a word stock ordering model training device, and aims to solve the problems that in the prior art, the time consumption for searching a link is prolonged, a user log of a third-party application needs to be relied on, and the searching ordering recommendation effect is poor.
In order to solve the above technical problem, the embodiments of the present specification are implemented as follows:
the embodiment of the specification provides a candidate word recommendation method, which comprises the following steps:
receiving a search request sent by a user, wherein the search request comprises search words;
selecting a first candidate word set to be recommended related to the search word from a first word bank according to the search word;
selecting a second candidate word set to be recommended related to the search word from at least one second word bank according to the search word;
sorting the first candidate word set to be recommended by adopting a first word stock sorting model to obtain a first sorting result;
sorting the second candidate word set to be recommended by adopting a second word stock sorting model to obtain a second sorting result; the second word bank ordering model is obtained by training search results returned by the training search words based on the training search words and other suggested word servers;
performing merging and sorting according to the first sorting result and the second sorting result;
and recommending the candidate words based on the combined and sorted result.
The embodiment of the specification provides a word stock ordering model training method, which comprises the following steps:
receiving search words carried in a search request sent by a user;
sending the search word and corresponding scene information when a user sends a search request to a suggested word server corresponding to a third-party application;
receiving a search result returned by a suggested word server of the third-party application, wherein the search result comprises a candidate word and a scoring result corresponding to the candidate word;
filtering the candidate words and storing the candidate words into a target word bank;
performing scoring learning training on the scoring result by using a preset machine learning algorithm to obtain a three-party lexicon ordering model;
the three-party word bank ordering model is used for ordering candidate words to be recommended selected from the target word bank.
An embodiment of the present specification provides a candidate word recommendation device, where the device includes:
the receiving module is used for receiving a search request sent by a user, wherein the search request comprises search terms;
the first selection module is used for selecting a first candidate word set to be recommended related to the search word from a first word bank according to the search word;
the second selection module is used for selecting a second candidate word set to be recommended related to the search word from at least one second word bank according to the search word;
the first sorting module is used for sorting the first candidate word set to be recommended by adopting a first word stock sorting model to obtain a first sorting result;
the second sorting module is used for sorting the second candidate word set to be recommended by adopting a second word stock sorting model to obtain a second sorting result; the second word bank ordering model is obtained by training search results returned by the training search words based on the training search words and other suggested word servers;
the merging module is used for merging and sorting according to the first sorting result and the second sorting result;
and the recommending module is used for recommending the candidate words based on the combined and sorted result.
An embodiment of the present specification provides a word stock ordering model training device, where the device includes:
the first receiving module is used for receiving search words carried in a search request sent by a user;
the sending module is used for sending the search word and corresponding scene information when the user sends the search request to a suggested word server corresponding to a third-party application;
the second receiving module is used for receiving a search result returned by a suggested word server of the third-party application, wherein the search result comprises candidate words and scoring results corresponding to the candidate words;
the filtering module is used for filtering the candidate words and storing the filtered candidate words into a target word bank;
the training module is used for performing scoring learning training on the scoring result by using a preset machine learning algorithm to obtain a three-party lexicon ordering model;
the three-party word bank ordering model is used for ordering candidate words to be recommended selected from the target word bank.
An electronic device provided in an embodiment of the present specification includes a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the candidate word recommendation method when executing the computer program.
An electronic device provided in an embodiment of the present specification includes a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the aforementioned method for training the lexicon order model when executing the computer program.
The embodiment of the specification adopts at least one technical scheme which can achieve the following beneficial effects:
receiving a search request sent by a user, wherein the search request comprises search terms; selecting a first candidate word set to be recommended related to the search word from a first word bank according to the search word; selecting a second candidate word set to be recommended related to the search word from at least one second word bank according to the search word; sorting the first candidate word set to be recommended by adopting a first word stock sorting model to obtain a first sorting result; sorting the second candidate word set to be recommended by adopting a second word stock sorting model to obtain a second sorting result; the second word bank ordering model is obtained by training search results returned by the training search words based on the training search words and other suggested word servers; merging and sorting according to the first sorting result and the second sorting result; and recommending the candidate words based on the combined and sorted result. Based on the scheme, the word bank built in the domain of the word bank and the three-party word bank (namely, the second word bank) can be directly called, and the model used for sequencing the three-party word bank is obtained by performing scoring learning training on the scoring result of the candidate words according to the third-party service, so that the time consumption of a link generated by calling the third-party service is avoided, the minimum search delay is ensured, the problem of invasion of user privacy caused by dependence on a user log applied by the third party is also avoided, and the effect of sequencing and recommending the candidate words is improved.
Drawings
In order to more clearly illustrate the embodiments of the present specification or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly introduced below, it is obvious that the drawings in the following description are only some embodiments described in the present specification, and for those skilled in the art, other drawings can be obtained according to the drawings without any creative effort.
Fig. 1 is a schematic diagram of an overall architecture of a platform involved in a practical application scenario according to the technical solution of the present specification;
fig. 2 is a schematic flowchart of a candidate word recommendation method provided in an embodiment of the present specification;
fig. 3 is a schematic flowchart of a method for training a lexicon ordering model according to an embodiment of the present disclosure;
fig. 4 is a schematic structural diagram of a candidate word recommending apparatus according to an embodiment of the present disclosure;
fig. 5 is a schematic structural diagram of a word stock ordering model training device provided in an embodiment of the present specification.
Detailed Description
In order to make those skilled in the art better understand the technical solutions in the present specification, the technical solutions in the embodiments of the present specification will be clearly and completely described below with reference to the drawings in the embodiments of the present specification, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any inventive step based on the embodiments of the present disclosure, shall fall within the scope of protection of the present application.
In the process that a user inputs a keyword through a search interface of an application program, a background service of the application program searches a word bank in the domain of the application program according to the keyword, and sorts the searched candidate words which are related to the keyword and then recommends the candidate words to the user, so that the user can select corresponding candidate words to search according to requirements; for example: when the user inputs the keyword "" game "", the suggested word server of the application program may give the following candidate suggested words "" game machine ", and" "game arcade".
When the search service of an application program also merges the deep search service of a third-party application, suggested words displayed above or below a search box in a search interface of the application program need to include not only suggested words searched by the application itself, but also search service results of the third-party application need to be displayed together. For example, the current search service of the application program a is further added with a deep search service of a third-party application B and a deep search service of a third-party application C, and when a user inputs a keyword through a search service page of the current application program a to perform a search, suggested words given by the current application program a need to be added with suggested word search results of the third-party application B and the third-party application C. For example, when the user searches for the word, "hangzhou" is a word, before accessing the third-party service, the suggested word gives "hangzhou subway taking code," hangzhou tong payment for treasure bus card "waiting result," hangzhou west lake tourist souvenir "and" hangbang dish "waiting result after accessing the third-party service (such as how to rob, what you get hungry, etc.) every day.
The existing candidate word recommendation modes fusing the third-party search service mainly comprise the following two modes, and the specific contents are as follows:
in the first mode, a request is directly sent to the third-party search service through the self application server, and then suggested words returned by the third-party search service are fused with suggested words determined by calling the self word bank. This approach can lead to the problem of long time-consuming link searching, and because of the directly introduced third-party service, the suggested word results returned by the third party cannot be handled, and the situation of inconsistent product positioning may occur.
In the second mode, a set of complete third-party word bank searching service is built in a server in the self-domain, the self-word bank and the third-party word bank in the self-application server are directly called, and the suggested word results obtained by searching are fused. The problem with this approach is that, first, the application itself lacks a valid user log of the third party application, and cannot learn this rule using a machine learning model, and the suggested word search ranking model cannot learn valid content. Second, if a third party user log is introduced, user privacy may be violated. In addition, the user intention of the third party is different from the application scene of the third party, and countless optimization can be performed on the third party service, so that the effect of the self-built third party suggested word ordering system is not as good as that of the third party service.
Aiming at the existing candidate word recommendation method fusing third-party services, in order to achieve the purpose that the fused candidate word ranking recommendation effect is better on the premise that the third-party services do not need to be directly called and the third-party user behaviors are not depended on, a candidate word recommendation method needs to be provided.
Fig. 1 is a schematic diagram of an overall architecture of a platform related to the technical solution of the present specification in an actual application scenario. The overall platform architecture may include a user end, a self application server, and a third-party application server, where the user end may be a mobile terminal used when a user logs in a current application, for example: mobile phones, tablet computers, and the like; the self application server can also be regarded as a background server corresponding to the application program currently logged in by the user and used for receiving the search request sent by the user side, and functions of a candidate word lexicon, a lexicon ordering model and the like are integrated in the self application server; the third-party application server can be regarded as a background server corresponding to other application programs except for the application of the third-party application server, and the suggested word service in the third-party application server is mainly used, so that the third-party application server can also be regarded as the third-party suggested word server. Although the third-party suggested word server also needs to be called in the embodiment of the description, the self application server uses the suggested word scoring result returned by the third-party suggested word server and trains the third-party word bank sequencing model in the self application server through machine learning, so that the third-party word bank sequencing model can absorb the sequencing characteristics of the third-party suggested word server, and the candidate words in the three-party word bank in the self application server are sequenced by using the scoring effect of the learned three-party model, so that the third-party service can be ensured not to be directly called, and the suggested word result returned by the third-party service and the suggested word in the self application server are fused and then recommended to the user, thereby prolonging the time consumption of searching links and avoiding the problem of controlling the suggested word result returned by the third party.
Based on the above-described scenarios, the following describes the embodiments of the present specification in detail.
Fig. 2 is a schematic flow chart of a candidate word recommendation method provided in an embodiment of this specification, where the method specifically includes the following steps:
step S210: receiving a search request sent by a user, wherein the search request comprises search words;
step S220: selecting a first candidate word set to be recommended related to the search word from a first word bank according to the search word;
step S230: selecting a second candidate word set to be recommended related to the search word from at least one second word bank according to the search word;
step S240: sorting the first candidate word set to be recommended by adopting a first word stock sorting model to obtain a first sorting result;
step S250: sorting the second candidate word set to be recommended by adopting a second word stock sorting model to obtain a second sorting result; the second word bank ordering model is obtained by training search results returned by the training search words based on the training search words and other suggested word servers;
step S260: performing merging and sorting according to the first sorting result and the second sorting result;
step S270: and recommending the candidate words based on the combined and sorted result.
The following describes in detail the implementation process of the candidate word recommendation method with reference to specific embodiments, and the specific contents are as follows:
in one or more embodiments of the present disclosure, a user may log in a current application program through a mobile terminal, and input a keyword that the user wants to query in a search box of a search page in a main interface of the current application program, at this time, the user side sends the keyword input by the user to a current application server. In practical applications, the term "user sends a search request" in the embodiments of the present specification refers to a search request sent by a user to a current application server through a mobile terminal.
The search terms contained in the search request, i.e. the incomplete words received by the input box during the input process of the user, are sent to the current application server, for example: the user wants to input the ant flower, and the keywords possibly received in the input process are m, ma, leech and the like.
Further, in this embodiment of the present specification, a search request sent when a user inputs a search term through a search page of a current application is received, where the search request further includes context information of the user when the user currently performs a search, and in an actual application, the context information of the user when searching includes, but is not limited to, the following: the geographic position information (such as longitude and latitude, location and the like) of the user and the time information when the user searches.
In one or more embodiments of the present specification, the first thesaurus may be considered as a thesaurus established by the current application according to the own service, and therefore, the first thesaurus may also be referred to as an own thesaurus, and candidate words in the first thesaurus are all candidate words generated by the current application (i.e., the own application). The second thesaurus can be regarded as a thesaurus created according to suggested words returned by the third-party service, so the second thesaurus can also be called a three-party thesaurus, and candidate words in the second thesaurus are all candidate words returned to the current application server by the third-party application. The first thesaurus and the second thesaurus may both be thesaurus established in the currently applied in-domain server, i.e. both the first thesaurus and the second thesaurus are in the currently applied in-domain server.
It should be noted that one third-party application may correspond to one third-party thesaurus (i.e. the second thesaurus), and therefore, when a plurality of third-party applications need to be fused, the number of the second thesaurus is not unique; of course, word banks corresponding to a plurality of third-party applications (i.e., three-party word banks) may be integrated into one second word bank, and any method is applicable to the present scheme.
Specifically, in the embodiment of the present specification, the operations of step S220 and step S230 described above may be performed in the following manner, specifically as follows:
according to the search words in the acquired search request, a first word bank and a second word bank are respectively called, some candidate words with higher correlation degree with the search words are selected from the first word bank and the second word bank, then the screened candidate words are respectively sorted by utilizing the sorting models corresponding to the respective word banks, namely, the candidate words selected from the first word bank are sorted in the first stage by utilizing the sorting model corresponding to the first word bank, and similarly, the candidate words selected from the second word bank are sorted in the first stage by utilizing the sorting model corresponding to the second word bank. In practical application, because the first word bank is a candidate word bank to which the first word bank is currently applied, the ranking model corresponding to the first word bank can be obtained by performing neural network training and learning on the scoring condition of the candidate words according to the application of the first word bank and a user log generated in an application scene of the first word bank, and a training process of the ranking model used by the word bank of the first word bank is not described in detail. The ranking model corresponding to the second lexicon is the three-party lexicon ranking model in the embodiment of the present specification, and the following will describe in detail the training process of the three-party lexicon ranking model.
After candidate words with high correlation degree with search words are selected from a first word bank and a second word bank respectively and are sorted in a first stage, all the candidate words sorted in the first stage are input into a merging algorithm, so that the candidate words are merged and sorted and recommended to a user.
In a specific embodiment, the merging algorithm is used to mix candidate words that are individually sorted in each lexicon (sorted in the first stage) together for reordering (which may be regarded as sorted in the second stage), but since the sorting models corresponding to the lexicons applied differently are not consistent and the scores in sorting may not be completely the score results of the same dimension, it is necessary to use a merging algorithm to mix the score results of the candidate words and the candidate words themselves together for reordering. In practical application, all the merging algorithms that can achieve the above-mentioned effects are applicable to this scheme, and certainly, besides the merging algorithm, the purpose of the above-mentioned second-stage sorting can be achieved by adopting other manners, for example, normalization processing.
With the content of the above embodiment, the main process of candidate word recommendation has been described, and a detailed description will be given below of how to obtain a ranking model corresponding to the second lexicon (i.e., a three-party lexicon ranking model) through training in actual operation.
In one or more embodiments of the present specification, an execution subject of the embodiments of the present specification may be a server corresponding to a current application (i.e., a self application), and therefore, after the self application server receives a search request sent by a user, the self application server may concurrently transmit search word and scene information in the search request to a third party suggested word server, so as to receive a search result, which is returned by the third party suggested word server and is generated by the third party application for the search word and the search scene of the user. It should be noted that the third-party suggested word server is not completely equal to the background server corresponding to the third-party application, and may be considered as a part of the application server.
The third-party suggested word server can be regarded as a server for providing suggested word search service by a third-party application, in the embodiment, the search service provided by the third-party application is referred to as the third-party service for short, the third-party service can be regarded as an application service with deep cooperation with an application party of the third-party service, and the third-party service is provided with an independent search engine and a search entrance, so that the third party can understand and optimize the search, and can train and learn a search ranking model of the third party service according to the ranking results of the search user, the query word and the candidate word of the third party service, that is, the search user of the third party provides rich search behaviors for the third party service to help the other party to optimize a search system of the third party.
In practical application, the search word and scene information in the search request are transmitted to the third-party suggested word server, therefore, the operation of receiving the search result generated by the third-party application on the search word and the search scene of the user and returned by the third-party suggested word server can be finished off-line or synchronously finished on-line, and particularly, after receiving the search request, the log of the search request can be directly sent to a third-party suggested word server, the self application server continues to search the first word stock and a plurality of second word stocks built in the self domain and returns the search result to the user normally, in the process, candidate words and scoring conditions returned by the third-party service are acquired offline to train a three-party ranking model of the third-party service, and the three-party ranking model is adjusted and updated, namely the process of offline processing; of course, after the log of the search request is sent to the third-party suggested word server, the candidate words and the scoring conditions returned by the third-party service are waited, a built-in three-party ranking model of the third-party service is trained on line, and then the candidate words selected from the third-party word bank are scored according to the three-party ranking model and then are merged and recommended to the user. The sequence numbers of the above steps in this specification do not constitute a limitation on the sequence of the implementation process of the present solution, and the present solution may be completed offline or completed online synchronously.
In one or more embodiments of the present disclosure, after receiving a candidate word result returned by a third-party service, the candidate word may be filtered according to a preset filtering logic meeting a current application requirement, and the filtered candidate word is stored in a second lexicon (i.e., a three-party lexicon). In practical applications, the filtering logic of the current application requirements may include candidate word filtering logic preset inside the current application, for example, sensitive words in the candidate words may be filtered, and the sensitive words may be words related to gambling, obscenities, drugs, violence, and the like. Candidate words returned by the third-party service are filtered by candidate word filtering logic set according to the requirements of the candidate words and then stored in the three-party word bank, so that the candidate words stored in the three-party word bank can meet the requirements of the candidate words.
Further, correspondingly, an embodiment of the present specification further provides a method for training a lexicon order model, as shown in fig. 3, which is a schematic flow diagram of the method for training a lexicon order model provided by the embodiment of the present specification, and the method specifically includes the following steps:
step S310: receiving search words carried in a search request sent by a user;
step S320: sending the search word and corresponding scene information when a user sends a search request to a suggested word server corresponding to a third-party application;
step S330: receiving a search result returned by a suggested word server of the third-party application, wherein the search result comprises a candidate word and a scoring result corresponding to the candidate word;
step S340: filtering the candidate words and storing the candidate words into a target word bank;
step S350: performing scoring learning training on the scoring result by using a preset machine learning algorithm to obtain a three-party lexicon ordering model;
the three-party word bank ordering model is used for ordering candidate words to be recommended selected from the target word bank.
The following describes in detail the implementation process of the word stock ranking model training method with reference to specific embodiments, and the specific contents are as follows:
in an embodiment of the present specification, the predetermined machine learning algorithm is a distillation algorithm, and the scoring learning training on the scoring result by using the predetermined machine learning algorithm may include the following:
obtaining the search word characteristics, the user characteristics and the search result characteristics determined by the current application according to the search request of the user, combining the candidate words, the scoring results corresponding to the candidate words, the search word characteristics, the user characteristics and the search result characteristics into the characteristics of model training, and performing scoring learning training operation on the characteristics of the model training by using a distillation algorithm.
Specifically, in the embodiment of the present specification, the distillation algorithm may be considered as a teaching algorithm, where the teaching algorithm refers to training a student model with inconsistent features by using the distillation algorithm, so that the student model can learn the scoring result of the third-party service, that is, the distillation algorithm refers to training a thesaurus ranking model applied by one party by using the scoring result of the candidate word of the other party, so that the thesaurus ranking model of the one party can obtain the same or almost the same scoring effect as the other party. Since the distillation algorithm itself is not improved by the scheme, the specific content of the distillation algorithm is not limited too much here. The following mainly describes model training features used in model training using the distillation algorithm, and how to learn the third-party scoring effect based on these features.
In this embodiment of the present specification, the features used in model training mainly include two parts, that is, a result returned by the third-party service and related features generated by the self-application, where the result returned by the third-party service is also a candidate word returned by the third-party service in the foregoing embodiment and a score of the candidate word, and the related features generated by the self-application include a search word feature, a user feature, a search result feature, and the like generated by a search system of the self-application according to a search request of a user. By abstracting the above features into features of model training and using these features to perform score learning training of models, the following describes a process of performing score learning training according to the above features with reference to a specific embodiment as follows:
for example, in a specific embodiment, the keyword searched by the user is "football", the third-party application B returns two search results, the first ranking is fire cube football No. 3 commodity, the corresponding scoring result is 0.2 point, the second ranking is lie football No. 5 commodity, and the corresponding scoring result is 0.1 point, so that the candidate word system applied by the candidate word system itself will combine some characteristics of the candidate word system itself according to the scoring results (such as fire cube football No. 3 commodity, 0.2 point; lie football commodity, No. 5 point, 0.1 point) returned by the other party, for example: the query term "football" corresponds to a statistical score (e.g., the number of searches per day), a classification of the query term, etc., as well as characteristics of the user (e.g., age, constellation, etc.), and characteristics of the search result (e.g., the returned search result is the search click rate/exposure rate of the fire cube football, etc.). And finally, combining the characteristics into model training characteristics, and training a three-party lexicon ordering model by using a distillation algorithm, so that the three-party lexicon ordering model can respectively play 0.2 point and 0.1 point for the fire cube brand football No. 3 commodity and the Lining No. 5 football commodity in the three-party lexicon.
Through the characteristics of the model training, the scoring result which is very similar to the third-party ranking scoring model can be learned according to the characteristics, although the trained three-party thesaurus ranking model can not necessarily print the same score as the third-party ranking scoring model, the problem that the privacy of a user is invaded due to the fact that the model is trained by relying on a user log of a third party is avoided, the ranking characteristics of the candidate words by the third-party service can be absorbed by learning the scoring result of the third-party service, and therefore the ranking effect of the candidate words in the three-party thesaurus in the self domain can be guaranteed to be better.
Further, in the embodiment of the present specification, after the distillation algorithm is used to perform the scoring learning training operation on the features of the model training, the trained model may be adjusted and learned according to the service scene log of the model, which is beneficial to making the candidate word to be recommended, which is finally scored by the three-party lexicon ranking model, more fit to the service of the model, and in practical application, the model may be adjusted and learned in the following manner, with specific contents as follows:
and adjusting, learning and training the three-party word bank ordering model after scoring learning training by using a preset regression model according to a service scene log generated after user search, and updating the original three-party word bank ordering model by using the three-party word bank ordering model after adjusting, learning and training.
In practical application, after the model (which may be considered as a three-party lexicon ranking model which is not yet adjusted and learned) is trained by using a distillation algorithm, the model can be adjusted by using the service data of the self scene, so that the scoring result of the three-party lexicon ranking model obtained through final adjustment is more fit with the self service scene. For example, when the user finally clicks the li ning football, the score of the model on the li ning football can be slightly increased, in practical application, a regression model of 0 and 1 can be adopted for adjustment, for example, when the user clicks the li ning football, the score of the model on the li ning football can be 1, and when the user does not click, the model can be scored as 0, in the actual adjustment training process, only a point needs to be adjusted every time, for example, the score on the li ning football is 0.3 at this time, but because the user does not click finally, the model requires the score on the li ning football to be 0, and through iteration of the sample, the score of the model on the li ning football can be changed into 0.29999 by the model.
In the embodiment of the present specification, the three-party thesaurus ranking model may be updated according to the three-party thesaurus ranking model adjusted and trained, where the updating may refer to directly replacing the original three-party thesaurus ranking model.
Based on the same idea, an embodiment of the present specification further provides a candidate word recommendation device, and as shown in fig. 4, a schematic structural diagram of the candidate word recommendation device provided in the embodiment of the present specification is provided, where the device 400 mainly includes:
a receiving module 401, configured to receive a search request sent by a user, where the search request includes a search term;
a first selecting module 402, configured to select, according to the search word, a first candidate word set to be recommended from a first word bank, where the first candidate word set is related to the search word;
a second selecting module 403, configured to select, according to the search word, a second candidate word set to be recommended from at least one second word bank, where the second candidate word set is related to the search word;
a first sorting module 404, configured to sort the first candidate word set to be recommended by using a first word bank sorting model, so as to obtain a first sorting result;
a second sorting module 405, configured to sort the second candidate word set to be recommended by using a second thesaurus sorting model, so as to obtain a second sorting result; the second word bank ordering model is obtained by training search results returned by the training search words based on the training search words and other suggested word servers;
a merging module 406, configured to perform merging and sorting according to the first sorting result and the second sorting result;
and the recommending module 407 is configured to recommend the candidate words based on the combined and sorted result.
Based on the same idea, an embodiment of the present specification further provides a device for training a lexicon ordering model, and as shown in fig. 5, the device 500 mainly includes:
a first receiving module 501, configured to receive a search term carried in a search request sent by a user;
a sending module 502, configured to send the search term and corresponding scene information when the user sends the search request to a suggested term server corresponding to a third-party application;
a second receiving module 503, configured to receive a search result returned by a suggested word server of the third-party application, where the search result includes a candidate word and a scoring result corresponding to the candidate word;
the filtering module 504 is configured to filter the candidate words and store the filtered candidate words in a target word bank;
a training module 505, configured to perform scoring learning training on the scoring result by using a predetermined machine learning algorithm, so as to obtain a three-party lexicon ordering model;
the three-party word bank ordering model is used for ordering candidate words to be recommended selected from the target word bank.
An embodiment of the present specification further provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the candidate word recommendation method when executing the computer program.
An embodiment of the present specification further provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the aforementioned method for training the lexicon order model when executing the computer program.
The foregoing description has been directed to specific embodiments of this disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the embodiments of the apparatus, the electronic device, and the nonvolatile computer storage medium, since they are substantially similar to the embodiments of the method, the description is simple, and the relevant points can be referred to the partial description of the embodiments of the method.
The apparatus, the electronic device, the nonvolatile computer storage medium and the method provided in the embodiments of the present description correspond to each other, and therefore, the apparatus, the electronic device, and the nonvolatile computer storage medium also have similar advantageous technical effects to the corresponding method.
In the 90 s of the 20 th century, improvements in a technology could clearly distinguish between improvements in hardware (e.g., improvements in circuit structures such as diodes, transistors, switches, etc.) and improvements in software (improvements in process flow). However, as technology advances, many of today's process flow improvements have been seen as direct improvements in hardware circuit architecture. Designers almost always obtain the corresponding hardware circuit structure by programming an improved method flow into the hardware circuit. Thus, it cannot be said that an improvement in the process flow cannot be realized by hardware physical modules. For example, a Programmable Logic Device (PLD), such as a Field Programmable Gate Array (FPGA), is an integrated circuit whose Logic functions are determined by programming the Device by a user. A digital system is "integrated" on a PLD by the designer's own programming without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Furthermore, nowadays, instead of manually manufacturing an integrated Circuit chip, such Programming is often implemented by "logic compiler" software, which is similar to a software compiler used in program development and writing, but the original code before compiling is also written by a specific Programming Language, which is called Hardware Description Language (HDL), and HDL is not only one but many, such as ABEL (Advanced Boolean Expression Language), AHDL (alternate Hardware Description Language), traffic, CUPL (core universal Programming Language), HDCal, jhddl (Java Hardware Description Language), Lava, Lola, HDL, PALASM, rhyd (Hardware Description Language), and the like, which are currently used in the field-Hardware Language. It will also be apparent to those skilled in the art that hardware circuitry that implements the logical method flows can be readily obtained by merely slightly programming the method flows into an integrated circuit using the hardware description languages described above.
The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic for the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller as pure computer readable program code, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may thus be considered a hardware component, and the means included therein for performing the various functions may also be considered as a structure within the hardware component. Or even means for performing the functions may be regarded as being both a software module for performing the method and a structure within a hardware component.
The systems, devices, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product with certain functions. One typical implementation device is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functionality of the various elements may be implemented in the same one or more software and/or hardware implementations in implementing one or more embodiments of the present description.
As will be appreciated by one skilled in the art, the present specification embodiments may be provided as a method, system, or computer program product. Accordingly, embodiments of the present description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and so forth) having computer-usable program code embodied therein.
The description has been presented with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the description. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
Computer-readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.
It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
This description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
The above description is only an example of the present specification, and is not intended to limit the present application. Various modifications and changes may occur to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.
Claims (16)
1. A candidate word recommendation method, the method comprising:
receiving a search request sent by a user, wherein the search request comprises search words;
selecting a first candidate word set to be recommended related to the search word from a first word bank according to the search word;
selecting a second candidate word set to be recommended related to the search word from at least one second word bank according to the search word;
sorting the first candidate word set to be recommended by adopting a first word stock sorting model to obtain a first sorting result;
sorting the second candidate word set to be recommended by adopting a second word stock sorting model to obtain a second sorting result; the second word bank ordering model is obtained by training search results returned by the training search words based on the training search words and other suggested word servers;
performing merging and sorting according to the first sorting result and the second sorting result;
and recommending the candidate words based on the combined and sorted result.
2. The method of claim 1, wherein receiving a search request sent by a user, the search request including a search term, comprises:
receiving a search request sent when a user inputs a search word through a search page of the current application, wherein the search request also comprises scene information of the user during searching.
3. The method according to claim 1, wherein before the sorting the second candidate word set to be recommended by using the second thesaurus sorting model, the method further comprises:
sending the search word and corresponding scene information when a user sends a search request to a suggested word server corresponding to a third-party application;
receiving a search result returned by a suggested word server of the third-party application, wherein the search result comprises a candidate word and a scoring result corresponding to the candidate word;
and performing scoring learning training on the scoring result by using a preset machine learning algorithm to obtain a three-party lexicon ordering model.
4. The method of claim 1, wherein the performing a merged ranking according to the first ranking result and the second ranking result, and performing candidate word recommendation based on the merged ranking result comprises:
and combining the first sorting result and the second sorting result by using a preset combination algorithm, then re-sorting, and sending the sorted candidate words to be recommended to the currently applied client so that the client can display the candidate words to be recommended to the user on a candidate word recommendation page.
5. The method of claim 3, further comprising:
and filtering the candidate words according to preset filtering logic meeting the current application requirements, and storing the filtered candidate words into the second word bank.
6. The method of claim 3, wherein the predetermined machine learning algorithm is a distillation algorithm, and wherein the scoring learning training of the scoring results using the predetermined machine learning algorithm comprises:
obtaining search word characteristics, user characteristics and search result characteristics determined by current application according to a user search request, combining the candidate words, scoring results corresponding to the candidate words, the search word characteristics, the user characteristics and the search result characteristics into model training characteristics, and performing scoring learning training operation on the model training characteristics by using the distillation algorithm.
7. The method of claim 6, after performing a scoring learning training operation on the model-trained features using the distillation algorithm, further comprising:
and adjusting, learning and training the three-party word bank ordering model after scoring learning training by using a preset regression model according to a service scene log generated after user search, and updating the original three-party word bank ordering model by using the three-party word bank ordering model after adjusting, learning and training.
8. The method of any of claims 1-7, wherein the first thesaurus is a thesaurus corresponding to a current application and the second thesaurus is a thesaurus corresponding to a third party application.
9. A method of training a thesaurus ranking model, the method comprising:
receiving search words carried in a search request sent by a user;
sending the search word and corresponding scene information when a user sends a search request to a suggested word server corresponding to a third-party application;
receiving a search result returned by a suggested word server of the third-party application, wherein the search result comprises a candidate word and a scoring result corresponding to the candidate word;
filtering the candidate words and storing the candidate words into a target word bank;
performing scoring learning training on the scoring result by using a preset machine learning algorithm to obtain a three-party lexicon ordering model;
the three-party word bank ordering model is used for ordering candidate words to be recommended selected from the target word bank.
10. A candidate word recommendation apparatus, the apparatus comprising:
the receiving module is used for receiving a search request sent by a user, wherein the search request comprises search terms;
the first selection module is used for selecting a first candidate word set to be recommended related to the search word from a first word bank according to the search word;
the second selection module is used for selecting a second candidate word set to be recommended related to the search word from at least one second word bank according to the search word;
the first sorting module is used for sorting the first candidate word set to be recommended by adopting a first word stock sorting model to obtain a first sorting result;
the second sorting module is used for sorting the second candidate word set to be recommended by adopting a second word stock sorting model to obtain a second sorting result; the second word bank ordering model is obtained by training search results returned by the training search words based on the training search words and other suggested word servers;
the merging module is used for merging and sorting according to the first sorting result and the second sorting result;
and the recommending module is used for recommending the candidate words based on the combined and sorted result.
11. The apparatus of claim 10, the second ordering module further to:
sending the search word and corresponding scene information when a user sends a search request to a suggested word server corresponding to a third-party application before the second candidate word set to be recommended is ranked by adopting a second word bank ranking model;
receiving a search result returned by a suggested word server of the third-party application, wherein the search result comprises a candidate word and a scoring result corresponding to the candidate word;
and performing scoring learning training on the scoring result by using a preset machine learning algorithm to obtain a three-party lexicon ordering model.
12. The apparatus of claim 11, the predetermined machine learning algorithm being a distillation algorithm, the second ranking module further to:
obtaining search word characteristics, user characteristics and search result characteristics determined by current application according to a user search request, combining the candidate words, scoring results corresponding to the candidate words, the search word characteristics, the user characteristics and the search result characteristics into model training characteristics, and performing scoring learning training operation on the model training characteristics by using the distillation algorithm.
13. The apparatus of claim 12, the second ordering module further to:
and after the distillation algorithm is used for performing scoring learning training operation on the characteristics of the model training, according to a service scene log generated after the user searches, adjusting, learning and training the three-party lexicon ordering model after the scoring learning training by using a preset regression model, and updating the original three-party lexicon ordering model by using the three-party lexicon ordering model after the adjusting, learning and training.
14. A thesaurus ranking model training apparatus, the apparatus comprising:
the first receiving module is used for receiving search words carried in a search request sent by a user;
the sending module is used for sending the search word and corresponding scene information when the user sends the search request to a suggested word server corresponding to a third-party application;
the second receiving module is used for receiving a search result returned by a suggested word server of the third-party application, wherein the search result comprises candidate words and scoring results corresponding to the candidate words;
the filtering module is used for filtering the candidate words and storing the filtered candidate words into a target word bank;
the training module is used for performing scoring learning training on the scoring result by using a preset machine learning algorithm to obtain a three-party lexicon ordering model;
the three-party word bank ordering model is used for ordering candidate words to be recommended selected from the target word bank.
15. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of any one of claims 1 to 8 when executing the program.
16. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of claim 9 when executing the program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010594333.7A CN111737595B (en) | 2020-06-24 | 2020-06-24 | Candidate word recommendation method, word bank ranking model training method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010594333.7A CN111737595B (en) | 2020-06-24 | 2020-06-24 | Candidate word recommendation method, word bank ranking model training method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111737595A true CN111737595A (en) | 2020-10-02 |
| CN111737595B CN111737595B (en) | 2024-02-06 |
Family
ID=72651241
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010594333.7A Active CN111737595B (en) | 2020-06-24 | 2020-06-24 | Candidate word recommendation method, word bank ranking model training method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111737595B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112364185A (en) * | 2020-11-23 | 2021-02-12 | 北京达佳互联信息技术有限公司 | Method and device for determining characteristics of multimedia resource, electronic equipment and storage medium |
| CN112883281A (en) * | 2021-03-29 | 2021-06-01 | 杭州同声相应科技有限公司 | User clustering search system based on deep learning |
| CN115314737A (en) * | 2021-05-06 | 2022-11-08 | 青岛聚看云科技有限公司 | Content display method, display equipment and server |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108319376A (en) * | 2017-12-29 | 2018-07-24 | 北京奇虎科技有限公司 | A kind of input association recommendation method and device that optimization business word is promoted |
| WO2019192261A1 (en) * | 2018-04-02 | 2019-10-10 | 阿里巴巴集团控股有限公司 | Payment mode recommendation method and device and equipment |
| US20190332867A1 (en) * | 2017-05-11 | 2019-10-31 | Tencent Technology (Shenzhen) Company Limited | Method and apparatus for retrieving similar video and storage medium |
| CN110532351A (en) * | 2019-07-19 | 2019-12-03 | 口口相传(北京)网络技术有限公司 | Recommend word methods of exhibiting, device, equipment and computer readable storage medium |
-
2020
- 2020-06-24 CN CN202010594333.7A patent/CN111737595B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190332867A1 (en) * | 2017-05-11 | 2019-10-31 | Tencent Technology (Shenzhen) Company Limited | Method and apparatus for retrieving similar video and storage medium |
| CN108319376A (en) * | 2017-12-29 | 2018-07-24 | 北京奇虎科技有限公司 | A kind of input association recommendation method and device that optimization business word is promoted |
| WO2019192261A1 (en) * | 2018-04-02 | 2019-10-10 | 阿里巴巴集团控股有限公司 | Payment mode recommendation method and device and equipment |
| CN110532351A (en) * | 2019-07-19 | 2019-12-03 | 口口相传(北京)网络技术有限公司 | Recommend word methods of exhibiting, device, equipment and computer readable storage medium |
Non-Patent Citations (2)
| Title |
|---|
| 刘慧婷;刘志中;王利利;吴信东;: "一般间隙序列模式挖掘的关键词抽取", 电子学报, no. 05 * |
| 郝胜男;赵领杰;: "一种基于ElasticSearch的推荐系统架构", 电脑知识与技术, no. 36 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112364185A (en) * | 2020-11-23 | 2021-02-12 | 北京达佳互联信息技术有限公司 | Method and device for determining characteristics of multimedia resource, electronic equipment and storage medium |
| CN112364185B (en) * | 2020-11-23 | 2024-02-06 | 北京达佳互联信息技术有限公司 | Method and device for determining characteristics of multimedia resources, electronic equipment and storage medium |
| CN112883281A (en) * | 2021-03-29 | 2021-06-01 | 杭州同声相应科技有限公司 | User clustering search system based on deep learning |
| CN115314737A (en) * | 2021-05-06 | 2022-11-08 | 青岛聚看云科技有限公司 | Content display method, display equipment and server |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111737595B (en) | 2024-02-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107346326B (en) | Method and system for information retrieval | |
| CN111737595B (en) | Candidate word recommendation method, word bank ranking model training method and device | |
| US9767183B2 (en) | Method and system for enhanced query term suggestion | |
| US20200372217A1 (en) | Method and apparatus for processing language based on trained network model | |
| CN114202370A (en) | Information recommendation method and device | |
| CN113688313A (en) | Training method of prediction model, information pushing method and device | |
| JP2021179979A (en) | Method for extracting attribute of item for shopping search | |
| CN113010640B (en) | Service execution method and device | |
| CN112733024A (en) | Information recommendation method and device | |
| CN113641894A (en) | Information recommendation method and device | |
| CN116186197A (en) | Topic recommendation method, device, electronic equipment and storage medium | |
| CN112699307A (en) | Information generation method and device | |
| CN111046304A (en) | Data searching method and device | |
| JP2021179980A (en) | Method for extracting category of item for shopping search | |
| CN112183584A (en) | Method and device for model training and business execution | |
| CN108595498B (en) | Question feedback method and device | |
| CN113343132B (en) | Model training method, information display method and device | |
| CN108595395B (en) | Nickname generation method, device and equipment | |
| CN113887234B (en) | Model training and recommending method and device | |
| CN110083687A (en) | A kind of information conversion method, equipment and storage medium | |
| CN113344078B (en) | Model training method and device | |
| CN113343085B (en) | Information recommendation method and device, storage medium and electronic equipment | |
| CN111461346B (en) | Network node characterization method, device and equipment | |
| CN114116816A (en) | Recommendation method and device | |
| KR102113663B1 (en) | Hierarchical classification-based incremental class learning method and computing device for digital storytelling |
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 |