CN112883281A - User clustering search system based on deep learning - Google Patents
User clustering search system based on deep learning Download PDFInfo
- Publication number
- CN112883281A CN112883281A CN202110336144.4A CN202110336144A CN112883281A CN 112883281 A CN112883281 A CN 112883281A CN 202110336144 A CN202110336144 A CN 202110336144A CN 112883281 A CN112883281 A CN 112883281A
- Authority
- CN
- China
- Prior art keywords
- search
- word
- user
- strategy
- information
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a user clustering search system based on deep learning, which comprises a search word bank, a search unit, a result display unit, a user unit and a historical database, wherein word information searched in a latest first time period is stored in the historical database; the search unit is configured with a search term matching strategy, and the search unit comprises a search tool. The invention provides a deep learning-based user clustering search system, which is used for presuming the search intention of a user by establishing a user characteristic model and adopting a collaborative filtering recommendation algorithm so as to improve the search experience of the user and quickly achieve the search purpose.
Description
Technical Field
The invention relates to the technical field of data analysis, in particular to a user clustering search system based on deep learning.
Background
Artificial intelligence (objective than human thinking) based on statistical principles: artificial intelligence refers to the science and technology of manufacturing machines with intelligence. Artificial intelligence based on statistical principles is the mainstream machine learning method at present, and the main difference between the artificial intelligence and other methods is that the mathematical basis is not logical and rule reasoning, but statistical criteria and principles. It derives a mathematical model of the problem world from a set of observations.
The behavior of internet users is as rich and complex as human thinking. In the process of data mining of the internet, keywords occupy a main position, and pages accessed by a user and contents liked by the user can be characterized by the keywords. The artificial intelligence based on the statistical principle provides basic ideas for the user behavior analysis of internet users in a search system.
Collective intelligence is a shared or group of intelligence that emerges from the cooperation and competition of many individuals, developing in bacterial, animal, human, and computer networks, and emerging in various forms of agreed-upon decision-making patterns. Collective intelligence also provides a basic solution to the problem in search systems for user behavior analysis of internet users.
In the current internet search system, it is desirable to improve the user search experience and achieve the purpose of user search quickly.
Disclosure of Invention
The invention aims to provide a user clustering search system based on deep learning, which is used for presuming the search intention of a user by establishing a user characteristic model and adopting a collaborative filtering recommendation algorithm so as to improve the search experience of the user and quickly achieve the search purpose.
In order to achieve the purpose, the invention provides the following technical scheme: a user clustering search system based on deep learning comprises a search word bank, a search unit, a result display unit, a user unit and a historical database, wherein word information searched in a first recent time period is stored in the historical database, the user unit comprises a plurality of different user characteristic models and a user model database, the user model database comprises an active word area, the active word area comprises a plurality of active words, and the active words are words of which the search times in a second recent time period are greater than a first value;
the search unit is configured with a search word matching strategy, the search unit comprises a search tool, the search tool comprises a search frame and a search starting item, the search word matching strategy comprises a candidate word sorting sub-strategy, the search word matching strategy is configured to acquire first search word information input in the search frame in real time, and call vocabulary information in a user model database, a history database or a search word library according to the candidate word sorting sub-strategy to generate a search word candidate entry, and the search word candidate entry comprises a plurality of search candidate words;
the search tool is provided with a first search strategy and a correlation contrast strategy, wherein the first search strategy comprises the steps of screening data containing first search word information in the active word area to generate a user intention search sub-word area, sequencing vocabulary information in the user intention search sub-word area according to the activity value from high to low, screening data containing the first search word information in the historical database to generate a historical intention search sub-word area, and sequencing the vocabulary information in the historical intention search sub-word area according to the activity value from high to low;
the relevancy contrast strategy is used for respectively calculating the relevancy between the vocabulary information in the user intention search sub-word area and the vocabulary information in the historical intention search sub-word area and the first search word information according to the sequence of the activity values from high to low and generating relevancy values;
the candidate word sorting sub-strategy comprises a first threshold, a second threshold, a third threshold and a fourth threshold, and is configured to display vocabulary information with an activity value larger than the first threshold and a correlation value larger than the second threshold in the user intention search sub-word region and vocabulary information with an activity value larger than the first threshold and a correlation value larger than the second threshold in the historical intention search sub-word region in the search word candidate entry in an interleaving manner.
Preferably, the result display unit includes a result item database, the result item database includes a plurality of result item information, the result display unit is configured with a second search strategy, the second search strategy includes that after a search starting item responds, a whole search word input in a search box at the time is acquired, and a search result data packet is generated according to the whole search word and the result item information matched with the whole search word in the result item database.
Preferably, the user model database further includes a typical vocabulary area and a high-weight vocabulary area, the high-weight vocabulary area includes a plurality of fixed high-weight vocabularies representing characteristics and habits of the user, the typical vocabulary area includes a plurality of typical vocabularies, and the typical vocabularies are obtained by calculation and screening according to the frequency and the times of occurrence of the high-weight vocabularies in the history database;
and the user unit comprises a user model construction strategy, and the user model construction strategy comprises calculation and classification according to the frequency and the times of the typical vocabulary and generation of a user characteristic model.
Preferably, the result display unit is further configured with a result display sorting policy, where the result display sorting policy includes sorting the result item information in the result data packet, which includes the typical vocabulary of the user, from high to low according to the activity of the included typical vocabulary, and randomly sorting the rest result item information.
Preferably, the user model building strategy comprises a fifth threshold, a third time period and a second numerical value, and the user model building strategy is configured to make the similarity of the typical vocabulary region larger than the fifth threshold and the same typical vocabulary searched in the latest third time period larger than the third threshold
The users with two values are divided into the same user characteristic model.
Preferably, the first search strategy further includes a group intention strategy, the group intention strategy includes obtaining active word areas of other users belonging to the same user feature model as the current user to form group active word areas, the first search strategy is further configured with data for screening information of the first search word in the group active word areas to generate group intention search sub-word areas, ranking vocabulary information in the group intention search sub-word areas according to activity values from high to low, and interspersing the vocabulary information with the activity value larger than a sixth threshold value in an original search word candidate entry to perform interval display.
Preferably, the search tool is further configured with an alternative search policy, the alternative search policy includes a group intention policy, the alternative search policy includes executing the group intention policy when a search candidate included in the search word candidate term generated according to the first search policy is smaller than a seventh threshold, the group intention policy includes obtaining active word regions of other users belonging to the same user feature model as the current user to form group active word regions, screening data including first search word information in the group active word regions to generate group intention search sub-term regions, and displaying vocabulary information in the group intention search sub-term regions after the original search word candidate term according to an order of an activity value from high to low.
Preferably, the search system is further configured with a data storage policy and a feature word bank, the feature word bank stores a plurality of initial feature words, the history database is provided with a plurality of data units corresponding to the initial feature words, the data storage policy includes acquiring a whole search word in a search box, comparing the whole search word with the initial feature words in the feature word bank, and storing the whole search word completely including the initial feature words in the data units corresponding to the initial feature words.
Preferably, the activity value can be characterized by calculating the number of times of occurrence of the vocabulary information in a fixed time period; the calculation method of the correlation value includes, but is not limited to, dividing the vocabulary information into different information degrees according to the predetermined category, grade, property and meaning.
Preferably, the feature word bank is configured with a feature word bank updating strategy and a data unit perfecting strategy, wherein the feature word bank updating strategy comprises that when an initial feature word matched with a whole search word does not exist in the feature word bank, a new data unit represented by the initial feature word is established; the data unit perfecting strategy comprises the step of refining a common intersection part of the vocabulary information in different data units into new initial characteristic words and dividing the new initial characteristic words into new data units.
Compared with the prior art, the invention has the beneficial effects that:
according to the invention, by establishing the user characteristic model, setting the active word area and generating the user intention search sub-word area through the active words, the latest search intention of the user can be obtained more quickly and accurately.
The search tool is provided with a group intention strategy, wherein the group intention strategy comprises the steps of obtaining active word areas of other users belonging to the same user characteristic model with a current user to form a group active word area; the collaborative filtering recommendation algorithm is adopted to predict the search intention of the users with the same preference, so that the search experience of the users can be improved, and the search purpose can be quickly achieved.
Drawings
FIG. 1 is a connection block diagram of a deep learning-based user clustering search system according to the present invention;
FIG. 2 is a schematic diagram illustrating candidate entry generation of search terms in a deep learning-based user clustering search system according to the present invention;
FIG. 3 is an explanatory diagram of a user model construction strategy in a deep learning-based user clustering search system according to the present invention;
FIG. 4 is an illustration of a data storage strategy in a deep learning-based user clustering search system according to the present invention;
FIG. 5 is an explanatory diagram of a user feature partitioning model in the deep learning-based user clustering search system according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, a first embodiment of the present invention provides a deep learning-based user cluster search system, which includes a search word bank, a search unit, a result display unit, a user unit and a history database, where the history database stores information of words searched in a first time period recently, the user unit includes a plurality of different user characteristic models and a user model database, the user model database includes an active word area, the active word area includes a plurality of active words, and the active words are words whose search frequency is greater than a first value in a second time period recently. According to the invention, by establishing the user characteristic model, setting the active word area and generating the user intention search sub-word area through the active words, the latest search intention of the user can be obtained more quickly and accurately.
As shown in fig. 1 and fig. 2, fig. 2 shows a schematic diagram of generating a candidate term of a search word, where a search unit is configured with a search word matching policy, the search unit includes a search tool, the search tool includes a search box and a search starting item, the search word matching policy includes a candidate word sorting sub-policy, the search word matching policy is configured to obtain first search word information input in the search box in real time, and invoke vocabulary information in a user model database, a history database or a search term database according to the candidate word sorting sub-policy to generate a candidate term of the search word, and the candidate term of the search word includes a plurality of search candidate words; the search tool is provided with a first search strategy and a correlation contrast strategy, wherein the first search strategy comprises the steps of screening data containing first search word information in the active word area to generate a user intention search sub-word area, sequencing vocabulary information in the user intention search sub-word area according to the activity value from high to low, screening data containing the first search word information in the historical database to generate a historical intention search sub-word area, and sequencing the vocabulary information in the historical intention search sub-word area according to the activity value from high to low; the relevancy contrast strategy is used for respectively calculating the relevancy between the vocabulary information in the user intention search sub-word area and the vocabulary information in the historical intention search sub-word area and the first search word information according to the sequence of the activity values from high to low and generating relevancy values; the candidate word sorting sub-strategy comprises a first threshold, a second threshold, a third threshold and a fourth threshold, and is configured to display vocabulary information with an activity value larger than the first threshold and a correlation value larger than the second threshold in the user intention search sub-word region and vocabulary information with an activity value larger than the first threshold and a correlation value larger than the second threshold in the historical intention search sub-word region in the search word candidate entry in a staggered manner; wherein the activity value can be characterized by calculating the occurrence times of the vocabulary information in a fixed time period; the calculation method of the correlation value includes, but is not limited to, dividing the vocabulary information into different information degrees according to the predetermined category, grade, property and meaning.
The first search strategy further comprises a group intention strategy, the group intention strategy comprises the steps of obtaining active word areas of other users belonging to the same user characteristic model with the current user to form a group active word area, screening data containing first search word information in the group active word area to generate a group intention search sub-word area, sorting the word information in the group intention search sub-word area according to the activity value from high to low, and inserting the word information with the activity value larger than a sixth threshold value into an original search word candidate entry to be displayed at intervals. The collaborative filtering recommendation algorithm is adopted to predict the search intention of the users with the same preference, so that the search experience of the users can be improved, and the search purpose can be quickly achieved.
Preferably, the result display unit includes a result item database, the result item database includes a plurality of result item information, the result display unit is configured with a second search strategy, the second search strategy includes that after a search starting item responds, a search whole word input in a search box at the moment is acquired, and a search result data packet is generated by screening the result item information matched with the search whole word in the result item database according to the search whole word; the result display unit is also provided with a result display sorting strategy, the result display sorting strategy comprises that the result item information containing the typical vocabulary of the user in the result data packet is sorted correspondingly according to the activity of the contained typical vocabulary from high to low, and the rest result item information is randomly arranged behind the result item information.
Preferably, the user model database further includes a typical vocabulary area and a high-weight vocabulary area, the high-weight vocabulary area includes a plurality of fixed high-weight vocabularies representing characteristics and habits of the user, the typical vocabulary area includes a plurality of typical vocabularies, and the typical vocabularies are obtained by calculation and screening according to the frequency and the times of occurrence of the high-weight vocabularies in the history database.
High-weight vocabularies may include, for example: speech, love, weight loss, camera shooting, students, europe and america, single body and the like can last for a certain time.
As shown in fig. 3, the subscriber unit includes a user model building strategy, which includes calculating and classifying according to the frequency and the times of the typical vocabulary occurrence and generating a user feature model, and corresponding labels can be built for different user feature models to be characterized. Such as high-grade white collar, enterprise high-line, teenagers, music enthusiasts, etc.
As shown in fig. 4, the search system is further configured with a data storage policy and a feature word library, where the feature word library stores a plurality of initial feature words, the history database is provided with a plurality of data units corresponding to the initial feature words, the data storage policy includes obtaining a whole search word in a search box, comparing the whole search word with the initial feature words in the feature word library, and storing the whole search word completely including the initial feature words in the data units corresponding to the initial feature words. Because the data units with the same initial characteristic words are set, the calculation time for searching the sub-word areas according to the historical intentions is reduced, and the searching efficiency and accuracy are improved.
As shown in fig. 5, the user model construction strategy includes a fifth threshold, a third time period and a second numerical value, and the user model construction strategy is configured to classify users with the similarity of the typical vocabulary region greater than the fifth threshold and the same typical vocabulary searched in the latest third time period greater than the second numerical value as the same user feature model; the feature word bank is configured with a feature word bank updating strategy and a data unit perfecting strategy, wherein the feature word bank updating strategy comprises the step of establishing a new data unit represented by an initial feature word when the initial feature word matched with the whole search word is not arranged in the feature word bank; the data unit perfecting strategy comprises the step of refining a common intersection part of the vocabulary information in different data units into new initial characteristic words and dividing the new initial characteristic words into new data units.
The method and the device help users to quickly search the information which the users want by predicting the search intention of the users with the same preference or habit.
Preferably, the search tool is further configured with an alternative search policy, the alternative search policy includes a group intention policy, the alternative search policy includes executing the group intention policy when a search candidate included in the search word candidate term generated according to the first search policy is smaller than a seventh threshold, the group intention policy includes obtaining active word regions of other users belonging to the same user feature model as the current user to form group active word regions, screening data including first search word information in the group active word regions to generate group intention search sub-term regions, and displaying vocabulary information in the group intention search sub-term regions after the original search word candidate term according to an order of an activity value from high to low.
The working principle is as follows: according to the method, the user characteristic model is established, the active word area is set, the user intention searching sub-word area is generated through the active words, the latest searching intention of the user can be obtained more quickly and accurately, and the active word area of other users belonging to the same user characteristic model with the current user is obtained through the group intention sub-strategy to form the group active word area; the collaborative filtering recommendation algorithm is adopted to predict the search intention of the users with the same preference, so that the search experience of the users can be improved, and the search purpose can be quickly achieved.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Claims (10)
1. A user clustering search system based on deep learning comprises a search word bank, a search unit and a result display unit, and is characterized by further comprising a user unit and a historical database, wherein word information searched in a first latest time period is stored in the historical database, the user unit comprises a plurality of different user characteristic models and user model databases, the user model database comprises an active word area, the active word area comprises a plurality of active words, and the active words are words of which the search times are greater than a first value in a second latest time period;
the search unit is configured with a search word matching strategy, the search unit comprises a search tool, the search tool comprises a search frame and a search starting item, the search word matching strategy comprises a candidate word sorting sub-strategy, the search word matching strategy is configured to acquire first search word information input in the search frame in real time, and call vocabulary information in a user model database, a history database or a search word library according to the candidate word sorting sub-strategy to generate a search word candidate entry, and the search word candidate entry comprises a plurality of search candidate words;
the search tool is provided with a first search strategy and a correlation contrast strategy, wherein the first search strategy comprises the steps of screening data containing first search word information in the active word area to generate a user intention search sub-word area, sequencing vocabulary information in the user intention search sub-word area according to the activity value from high to low, screening data containing the first search word information in the historical database to generate a historical intention search sub-word area, and sequencing the vocabulary information in the historical intention search sub-word area according to the activity value from high to low;
the relevancy contrast strategy is used for respectively calculating the relevancy between the vocabulary information in the user intention search sub-word area and the vocabulary information in the historical intention search sub-word area and the first search word information according to the sequence of the activity values from high to low and generating relevancy values;
the candidate word sorting sub-strategy comprises a first threshold, a second threshold, a third threshold and a fourth threshold, and is configured to display vocabulary information with an activity value larger than the first threshold and a correlation value larger than the second threshold in the user intention search sub-word region and vocabulary information with an activity value larger than the first threshold and a correlation value larger than the second threshold in the historical intention search sub-word region in the search word candidate entry in an interleaving manner.
2. The deep learning based user cluster search system according to claim 1, wherein: the result display unit comprises a result item database, the result item database comprises a plurality of result item information, the result display unit is configured with a second search strategy, the second search strategy comprises the steps of obtaining a search whole word input in a search frame at the moment after a search starting item responds, and screening the result item information matched with the search whole word in the result item database to generate a search result data packet according to the search whole word.
3. The deep learning based user cluster search system according to claim 2, wherein: the user model database also comprises a typical vocabulary area and a high-weight vocabulary area, wherein the high-weight vocabulary area comprises a plurality of fixed high-weight vocabularies representing the characteristics and habits of a user, the typical vocabulary area comprises a plurality of typical vocabularies, and the typical vocabularies are obtained by calculation and screening according to the frequency and the frequency of the occurrence of the high-weight vocabularies in the history database;
and the user unit comprises a user model construction strategy, and the user model construction strategy comprises calculation and classification according to the frequency and the times of the typical vocabulary and generation of a user characteristic model.
4. The deep learning based user cluster search system of claim 3, wherein: the result display unit is also provided with a result display sorting strategy, the result display sorting strategy comprises that the result item information containing the typical vocabulary of the user in the result data packet is sorted correspondingly according to the activity of the contained typical vocabulary from high to low, and the rest result item information is randomly arranged behind the result item information.
5. The deep learning based user cluster search system of claim 4, wherein: the user model building strategy comprises a fifth threshold, a third time period and a second numerical value, and the user model building strategy is configured to divide users with the similarity of the typical vocabulary region larger than the fifth threshold and the same typical vocabulary searched in the latest third time period larger than the second numerical value into the same user feature model.
6. The deep learning based user cluster search system of claim 5, wherein: the first search strategy further comprises a group intention strategy, the group intention strategy comprises the steps of obtaining active word areas of other users belonging to the same user characteristic model with the current user to form a group active word area, screening data containing first search word information in the group active word area to generate a group intention search sub-word area, sorting the word information in the group intention search sub-word area according to the activity value from high to low, and inserting the word information with the activity value larger than a sixth threshold value into an original search word candidate entry to be displayed at intervals.
7. The deep learning based user cluster search system of claim 5, wherein: the search tool is further configured with alternative search strategies, the alternative search strategies comprise group intention strategies, the group intention strategies comprise executing the group intention strategies when the search candidate words contained in the search word candidate entries generated according to the first search strategy are smaller than a seventh threshold value, the group intention strategies comprise obtaining active word areas of other users belonging to the same user characteristic model as the current user to form group active word areas, screening data containing first search word information in the group active word areas to generate group intention search sub-word areas, and displaying vocabulary information in the group intention search sub-word areas after the original search word candidate entries according to the sequence of the activity degree from high to low.
8. The deep learning based user cluster search system according to claim 6 or 7, wherein: the search system is further provided with a data storage strategy and a feature word bank, wherein a plurality of initial feature words are stored in the feature word bank, a plurality of data units corresponding to the initial feature words are arranged in the history database, the data storage strategy comprises the steps of obtaining the whole search words in the search frame, comparing the whole search words with the initial feature words in the feature word bank, and storing the whole search words completely containing the initial feature words in the data units corresponding to the initial feature words.
9. The deep learning based user cluster search system of claim 8, wherein: the activity value can be characterized by calculating the occurrence times of the vocabulary information in a fixed time period; the calculation method of the correlation value includes, but is not limited to, dividing the vocabulary information into different information degrees according to the predetermined category, grade, property and meaning.
10. The deep learning based user cluster search system of claim 9, wherein: the feature word bank is configured with a feature word bank updating strategy and a data unit perfecting strategy, wherein the feature word bank updating strategy comprises the step of establishing a new data unit represented by an initial feature word when the initial feature word matched with the whole search word is not arranged in the feature word bank; the data unit perfecting strategy comprises the step of refining a common intersection part of the vocabulary information in different data units into new initial characteristic words and dividing the new initial characteristic words into new data units.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110336144.4A CN112883281B (en) | 2021-03-29 | 2021-03-29 | User clustering search system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110336144.4A CN112883281B (en) | 2021-03-29 | 2021-03-29 | User clustering search system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112883281A true CN112883281A (en) | 2021-06-01 |
| CN112883281B CN112883281B (en) | 2022-10-14 |
Family
ID=76039780
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110336144.4A Active CN112883281B (en) | 2021-03-29 | 2021-03-29 | User clustering search system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112883281B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170220675A1 (en) * | 2016-01-29 | 2017-08-03 | International Business Machines Corporation | Topic generation for a publication |
| CN107730343A (en) * | 2017-09-15 | 2018-02-23 | 广州唯品会研究院有限公司 | A kind of user's merchandise news method for pushing and equipment based on picture attribute extraction |
| CN107798066A (en) * | 2017-09-25 | 2018-03-13 | 北京小度信息科技有限公司 | A kind of search term method for pushing, device and terminal |
| CN111737595A (en) * | 2020-06-24 | 2020-10-02 | 支付宝(杭州)信息技术有限公司 | Candidate word recommendation method, word stock ordering model training method and device |
-
2021
- 2021-03-29 CN CN202110336144.4A patent/CN112883281B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170220675A1 (en) * | 2016-01-29 | 2017-08-03 | International Business Machines Corporation | Topic generation for a publication |
| CN107730343A (en) * | 2017-09-15 | 2018-02-23 | 广州唯品会研究院有限公司 | A kind of user's merchandise news method for pushing and equipment based on picture attribute extraction |
| CN107798066A (en) * | 2017-09-25 | 2018-03-13 | 北京小度信息科技有限公司 | A kind of search term method for pushing, device and terminal |
| CN111737595A (en) * | 2020-06-24 | 2020-10-02 | 支付宝(杭州)信息技术有限公司 | Candidate word recommendation method, word stock ordering model training method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112883281B (en) | 2022-10-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112632385B (en) | Course recommendation method, course recommendation device, computer equipment and medium | |
| US20200184275A1 (en) | Method and system for generating and correcting classification models | |
| CN109829104B (en) | Semantic similarity based pseudo-correlation feedback model information retrieval method and system | |
| CN111708873A (en) | Intelligent question answering method and device, computer equipment and storage medium | |
| WO2021093755A1 (en) | Matching method and apparatus for questions, and reply method and apparatus for questions | |
| CN110209808B (en) | Event generation method based on text information and related device | |
| US8150822B2 (en) | On-line iterative multistage search engine with text categorization and supervised learning | |
| US9183285B1 (en) | Data clustering system and methods | |
| CN109299271A (en) | Training sample generation, text data, public sentiment event category method and relevant device | |
| Shi et al. | Hashtagger+: Efficient high-coverage social tagging of streaming news | |
| CN111444344A (en) | Entity classification method, entity classification device, computer equipment and storage medium | |
| CN116455861B (en) | Big data-based computer network security monitoring system and method | |
| CN111831924A (en) | Content recommendation method, device, equipment and readable storage medium | |
| CN112463944A (en) | Retrieval type intelligent question-answering method and device based on multi-model fusion | |
| CN116628339A (en) | Educational resource recommendation method and system based on artificial intelligence | |
| CN106095941B (en) | Big data knowledge base-based solution recommendation method and system | |
| CN113868406A (en) | Search method, search system, and computer-readable storage medium | |
| CN116882414B (en) | Automatic comment generation method and related device based on large-scale language model | |
| CN111767404B (en) | Event mining method and device | |
| JPH07129593A (en) | Text selecting device | |
| CN112883281B (en) | User clustering search system based on deep learning | |
| CN115827990A (en) | Searching method and device | |
| CN109254993B (en) | Text-based character data analysis method and system | |
| Boryczka et al. | Using Differential Evolution in order to create a personalized list of recommended items | |
| Zhou et al. | Research on personalized e-learning based on decision tree and RETE algorithm |
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 |