CN109582868A - The search recommended method of preference is clicked based on term vector weighting, support vector regression and user - Google Patents
The search recommended method of preference is clicked based on term vector weighting, support vector regression and user Download PDFInfo
- Publication number
- CN109582868A CN109582868A CN201811422626.6A CN201811422626A CN109582868A CN 109582868 A CN109582868 A CN 109582868A CN 201811422626 A CN201811422626 A CN 201811422626A CN 109582868 A CN109582868 A CN 109582868A
- Authority
- CN
- China
- Prior art keywords
- text
- search
- user
- vector
- preference
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The present invention relates to a kind of search recommended methods that preference is clicked based on term vector weighted average, support vector regression and user.Its summary of the invention mainly includes that (1) proposes a kind of search recommended models that preference is clicked based on search prefix semantic dependency and user;(2) the semantic dependency calculation method of a kind of the search prefix text insertion based on support vector regression and content text insertion is proposed;(3) a kind of weight modification method that preference sum is clicked based on transfer learning and user is proposed.
Description
Technical field
The present invention relates to computer application technologies, and one kind is based on term vector weighted average, support vector regression and use
The search recommended method of family click preference.
Background technique
In recent years, due to the fast development of computer field, the information content that people obtain is exponentially increased.In face of from four sides
From all directions to the information to come tumbling, people often feel dazzled.On the other hand, artificial intelligence technology is constantly broken through
And progress, the actual queries that people have new demand to meet user as far as possible existing search engine are intended to.Therefore, such as
What combines the artificial intelligence technology information needed for obtaining user in mass data quick, intelligently, so that improving search recommends system
The service quality of system is a major challenge in present age computer search recommendation field.
Huge and mixed and disorderly feature is showed based on Internet resources, two kinds of retrieval techniques occur: sort-type is retrieved and searched
Cable-styled retrieval.Key search can be used when user is retrieved using classified catalogue, or go to browse using catalog classification.Search
Then by the way of information inquiry, the inquiry request that search engine receives user can return the result list for formula retrieval.Both
Technology reduces the difficulty of user search information to a certain extent, alleviates the pressure that people obtain information from mass data
Power, but there is no according to user the characteristics of, use the relevant informations such as background provide be suitble to user's individual resource retrieval recommendation
Service, cannot fully meet user demand.On the other hand, also because return information in junk information enormous amount, search
Index holds up the demand that can not fully meet user to INFORMATION DISCOVERY.In this background, recommender system is come into being, it can connect
User and information are met, value is created.
Search recommended technology refers to that search engine submits the keyword in query word or query word using certain according to user
Algorithm, analyze user's underlying search intention, recommend after several inquiries of user are recommended and select the technology of word.Draw with simple search
It holds up and compares, search for recommended engine at present and have been provided with preliminary intelligence: obtaining the interest of user by excavating user's search log
The information such as hobby establish user interest model, so that returning to user may interested information;Or pass through word in ontology
Between semantic meaning or objective reality the information carriers such as text, video, excavate the real semantic association between information,
And user is recommended by the association size between them;Or the mode based on social networks recommends relevant information;Or it is logical
The similitude for crossing research query word does query expansion, inquiry cluster and inquiry similitude.But search process is likely to occur part
User can not know to entirely accurate the information that oneself is needed or be difficult with simple keyword come the case where stating demand, example
Such as, when user is investigated in a unfamiliar field, they want to obtain more knowledge to deepen to the field
Cognition, rather than purposefully look for a certain category information;Alternatively, there are specific search intentions by user, but it is limited to have and knows
Know and limited does not know how to express.At this time, it may be necessary to show more has the associated information of reality to user with current search, and cannot
Only show the information of user model.On the other hand, for cumbersome search result, junk information how is further reduced again not
The careless omission possible search intention of user, it is necessary to which search is recommended to be further optimized.
Nowadays with computer, internet, machine learning development, promote artificial intelligence field to develop rapidly.It may be said that
Where where is it computer and data, where is it artificial intelligence just have the theory using artificial intelligence, methods and techniques.Machine
Study be realize artificial intelligence a kind of method, in order to make search for recommender system it is more intelligent, the present invention by user click preference with
Natural language processing technique and the method for machine learning combine, and analyze data using algorithm, therefrom learn and make deduction or pre-
It surveys.
Summary of the invention
The invention proposes a kind of search for clicking preference based on term vector weighted average, support vector regression and user to push away
Method is recommended, mainly includes three big contents:
(1) a kind of search recommended models that preference is clicked based on search prefix semantic dependency and user are proposed;
(2) propose that a kind of search prefix text insertion based on support vector regression is related to the semanteme that content text is embedded in
Property calculation method;
(3) a kind of weight modification method that preference sum is clicked based on transfer learning and user is proposed.
Particular content is as follows:
(1) a kind of search recommended models that preference is clicked based on search prefix semantic dependency and user: the recommendation are proposed
The flow chart of model is as shown in Figure 1, the semantic dependency and user that it comprehensively considers search prefix and content text are clicked partially
Recommend to do well.
Step 1: consider the semantic dependency of search prefix and content text.It is user's lookup that user, which inputs search content,
The prefix of content title text, and a search content corresponds to multiple possible users and searches content title text.For search
Content text and its corresponding each content title text are weighted and averaged based on term vector, construct search content text and its right
The text vector for each content title text answered.Semantic dependency model is established using support vector regression algorithm again and calculates this
The relevance values of two texts.The input of model is the text vector of two class texts, and output is the relevance values of two class texts.It builds
The process of mould will be described in detail in content 2.
Step 2: the semantic dependency model of preference amendment first part is clicked in conjunction with user.This is based partially on first
Partial semantic dependency simultaneously combines user to click preference building model.Because user is emerging to the sense of different content title texts
Interesting degree is different, and the present invention counts the clicking rate of different content title text, and the click preference of user is measured with clicking rate.
Search content text vector, content title text vector and clicking rate are built based on transfer learning and support vector regression
Mould, when modeling, needs the model initialization for obtaining first part to be embedded in, to achieve the effect that correction model.When finally restraining
Model be final search recommended models.Specific modeling process is described in detail in content 3.
(2) propose that a kind of search prefix text insertion based on support vector regression is related to the semanteme that content text is embedded in
Property calculation method.
Definition: M is data set, M=S, S ' }, S '={ s1’,s2’,...,sn', S represents the search text of user's input
This, S ', which is represented, inquires content title text, such textual data one according to the complete demand of user of the search text prediction of input
Shared n, be s respectively1’-sn’。
The semantic dependency calculation method includes two steps:
Step 1: for search text and content text using Chinese words segmentation, term vector constructing technology and weighting skill
Art is converted to text vector, and process is as follows:
1) segmented to text using Chinese words segmentation: the present invention segments to obtain search text and interior using jieba
Hold the word sequence of title text.S={ b is obtained after participle1’,b2’…bi', s1'={ b11’,b12’…b1j’},s2’-sn' obtain
The same s of form1’。b1’-bi' represent i word after search text S participle, b11’-b1j' represent content title text s1' after participle
J word.
2) text is pre-processed: judges whether there is some additional characters and meaningless word in text, if there is then
It needs to carry out text the processing of stop words, otherwise skips this step.
3) word is converted to term vector using term vector constructing technology: the present invention is for search text and content text
Word sequence uses word2vec term vector constructing technology, obtains the corresponding term vector of each word.Total word number that the present invention is included
It is 25000, term vector dimension is 100 dimensions, obtains S={ v after building term vector1,v2…vi},s1'={ v11’,v12’…
v1j', s2’-sn' the obtained same s of form1'.Wherein v1-viThe corresponding vector of i word of search text S is represented, each vector is
100 dimensions, v11’-v1j' represent content title text s1' the corresponding vector of j word, each vector is 100 dimensions.
4) word-based vector sum weighting technique constructs text vector: the power of each word in text is calculated using tfidf technology
Weight, the process for calculating weight are as follows:
With the word b searched in text1' for, calculating word b first1' frequency that occurs in current search text.Formula is such as
Under:
Wherein molecule is word b1' number in current text is appeared in, denominator is the appearance of all words in the text
The sum of number.
Then reverse document-frequency is calculated, it is the measurement of a word general importance.Formula is as follows:
Wherein molecule is all search text numbers, and denominator is comprising word b1Search text number.
Finally (1) formula is multiplied with the result of (2) formula can be obtained the weight of each word, average weighted using vector
Mode is superimposed the term vector in text, if the corresponding vector of search text is d,
Wherein to be that all words are corresponding in text sum molecule multiplied by respective tf, idf, and denominator is all words hereof
The sum of frequency of occurrence.Obtain S={ d after text vectors, dt, dsRepresent the text vector of search text, dtRepresent inquiry
Content title text vector.
Step 2: using support vector regression algorithm, and repetitive exercise calculates the semantic dependency of two text vectors.Meter
The process for calculating relevance values is as follows:
1) it models: given training sample D={ (dt1,ds1,y1),(dt2,ds2,y2)…(dtm,dsm,ym), wherein ds1-dsm
It is m search text vector and dt1-dtmIt is m content text vector, y1-ymIt is that search text is related to content title text
Property value, the inquiry content title text relevant value clicked be 1, the content title text relevant value that do not clicked be 0.
The present invention allows the point in each training set to be fitted to a hyperplane as far as possible, and what which represented is search text, content mark
Inscribe the model f of relationship between text and correlation1(dt1,ds1)1=w1 Tφ(dt1,ds1)+b1, f1(dt1,ds1) it is quasi- according to model
The relevance values of the search text and content title text of conjunction, w1The weight of representative model.It sets f (X)1With true correlation value
Between be up to ∈ deviation, obtain two foreign peoples's supporting vector f1(dt1,ds1)-ε and f1(dt1,ds1)+ε.SoIt is
Two foreign peoples's supporting vectors work as f to the sum of the distance of hyperplane1(dt1,ds1) and true correlation value between difference greater than ∈ count
Calculate loss.Shown in objective function such as formula (4):
This is a convex quadratic programming problem, directly can calculate packet with ready-made optimization and solve.Find out w1,b1Value obtain
Semantic dependency regression model and preservation model.
2) after modeling is completed, search text vector and content text vector to be calculated are inputted, obtained output valve is just
It is the relevance values for searching for text vector and content text vector.
(3) a kind of weight modification method that preference sum is clicked based on transfer learning and user is proposed: since user is to difference
The interest level of content is different, so the correlation of search text and content title text needs to consider user preference, unites
The click probability that different content text is clicked when counting out the same search prefix of user's input, search text is represented with clicking rate
With the correlation of content title text.Training set the N={ (d of new modelt1,ds1,y1’),(dt1,ds1,y2’)…(dt1,ds1,
ym'), wherein ds1-dsmIt is m search text vector, dt1-dtmIt is m content text vector, y1’-ym' it is different content text
Click probability.Model initialization obtained in content 2 is embedded into support vector regression model by the thought based on transfer learning
In, it substitutes into as search text vector and the content text vector of input value and as the click probability of output valve, obtains
f2(dt1,ds1,w1)=w2 Tφ(dt1,ds1)+b2 (5)
The further repetitive exercise model, algorithm solution procedure are same as above.The model is simultaneous while retaining semantic dependency
User behavior is cared for, to achieve the purpose that correct regressive prediction model weight using user behavior.
After obtaining new model, user is searched for text to the present invention and possible content title text is converted into text vector
After input, the relevance values based on the available search text and content text of new model saved.Relevance values at this time
It is to combine the result that semantic dependency and user's click preference factor obtain.It is finally sorted according to relevance values, exports TOPN
Corresponding content title text is as the content finally recommended.
Detailed description of the invention
Fig. 1 is work flow diagram of the present invention.
Specific embodiment
The present invention is that the search recommended method of preference is clicked based on term vector weighted average, support vector regression and user.
Development language is python, and exploitation environment is win10, sample data: register { " what registered mail is ", " net of registering
Upper reservation ":, " register and net official website ":, " registered mail ":, " registering ", " platform of registering ", " net of registering " } net of registering.User's input
Searching for text is to register, and the inquiry content title text finally clicked is net of registering.Specific step is as follows:
Step 1: data preparation
Be the form storage needed by Data Format Transform: a search text has multiple possible content title texts
This, but the content title text that user clicked is a part therein.First search text and content title text with
And whether click the content text and be mapped, relevance values were 1 if clicking, and relevance values were 0 if click.
Step 2: Chinese word segmentation and Text Pretreatment
First segmented using the jieba segmenter of open source.Search text and inquiry content title text, output are
Search for text word corresponding with inquiry content title text.Download common Chinese stoplist again, judge in text whether include
Stop words has, and removes stop words, otherwise carries out in next step.
Step 3: building term vector
All search texts first are inputted using term vector tool word2vec and inquiry content title text loads corpus, so
Corpus is trained afterwards to obtain term vector model as training sample.Next each word is traversed, is obtained using model each
The term vector of word.Input is search text word corresponding with inquiry content title text, and output is the corresponding vector of each word.
Step 4: building search prefix text vector and content title text vector
Using TF_IDF weighting technique search text word corresponding with inquiry content title text, each word is obtained
Weighted value.One text is divided into one or more words, then each word directed quantity and weight, again remove the addition of vector sum weight
With the number of the word in a word, just obtains search text vector and inquire content title text vector.
Step 5: text relevant is calculated by support vector regression
First using support vector regression algorithm training initial relevance model, parameter is search text vector and inquiry content
Title text vector, if click, obtain model 1.The clicking rate for counting each inquiry content title text again, at the beginning of model 1
Beginningization is embedded into support vector regression model, and the further repetitive exercise model obtains revised model.It finally calls and repairs
Model after just, search text vector sum inquires content title text vector, the corresponding content mark of TOPN of output valve
Text is inscribed as last recommendation.
Claims (4)
1. a kind of search recommended method for clicking preference based on term vector weighted average, support vector regression and user, feature
Include:
1) a kind of search recommended models that preference is clicked based on search prefix semantic dependency and user are proposed;
2) it proposes a kind of based on the semantic dependency of the insertion of the search prefix text of support vector regression and content text insertion
Calculation method;
3) a kind of weight modification method that preference sum is clicked based on transfer learning and user is proposed.
2. the search according to claim 1 for clicking preference based on term vector weighted average, support vector regression and user
Recommended method, it is characterised in that: the semantic dependency and user for comprehensively considering search prefix and content text click preference,
Realize that search is recommended;The search recommended models are related to two parts, and first part is search prefix semantic dependency: user's input is searched
Rope content is the prefix that user searches content title text, searches for text with content title text and semantically there is correlation;
It uses for search text and content title text and is characterized based on the average weighted text vector constructing technology of term vector, adopted
The semantic dependency of the two is measured out with support vector regression algorithm;Second part is to combine user to click preference to correct first
The semantic dependency model divided, to achieve the purpose that take into account semantic dependency and click preference: counting user is in same search
To the click probability of different texts under prefix, click the click preference that probability then represents user, based on transfer learning and support to
Amount is returned and is further modified to semantic dependency model;Model when finally restraining is final search recommended models.
3. the search according to claim 1 for clicking preference based on term vector weighted average, support vector regression and user
Recommended method, it is characterised in that: using Chinese words segmentation processing search text and content text, form search text and content
The word sequence of text;Term vector is constructed using word2vec term vector technology, and each word in text is calculated using tfidf technology
Weight, the term vector in text is superimposed using the average weighted mode of vector, to form text vector;With search for text to
Amount and content text vector are as input, and using support vector regression algorithm, repetitive exercise calculates the language of a text vector two-by-two
Adopted correlation.
4. the search according to claim 1 for clicking preference based on term vector weighted average, support vector regression and user
Recommended method, it is characterised in that: click preference for user, clicked not when counting the same search prefix of user's input first
With the click probability of content text;Further according to semantic dependency regression model obtained in the 2nd point feature, based on transfer learning
Thought take the model as the support vector regression model of initialization, is defeated to search for prefix text vector and content text vector
Enter, is output, the further repetitive exercise model with the corresponding click probability of content text;Obtained new model is finally restrained,
While retaining semantic dependency, user behavior is taken into account, utilizes user behavior amendment regressive prediction model weight to reach
Purpose, to realize more accurate recommendation.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811422626.6A CN109582868A (en) | 2018-11-27 | 2018-11-27 | The search recommended method of preference is clicked based on term vector weighting, support vector regression and user |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811422626.6A CN109582868A (en) | 2018-11-27 | 2018-11-27 | The search recommended method of preference is clicked based on term vector weighting, support vector regression and user |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109582868A true CN109582868A (en) | 2019-04-05 |
Family
ID=65924832
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811422626.6A Pending CN109582868A (en) | 2018-11-27 | 2018-11-27 | The search recommended method of preference is clicked based on term vector weighting, support vector regression and user |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109582868A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111274808A (en) * | 2020-02-11 | 2020-06-12 | 支付宝(杭州)信息技术有限公司 | Text retrieval method, model training method, text retrieval device, and storage medium |
CN111782871A (en) * | 2020-06-18 | 2020-10-16 | 湖南大学 | Cross-modal video time positioning method based on space-time reinforcement learning |
CN112487274A (en) * | 2020-12-02 | 2021-03-12 | 重庆邮电大学 | Search result recommendation method and system based on text click rate |
EP4145308A4 (en) * | 2020-05-19 | 2023-09-20 | Huawei Technologies Co., Ltd. | Search recommendation model training method, and search result sorting method and device |
CN111782871B (en) * | 2020-06-18 | 2024-06-25 | 湖南大学 | Cross-modal video time positioning method based on space-time reinforcement learning |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110184806A1 (en) * | 2010-01-27 | 2011-07-28 | Ye Chen | Probabilistic recommendation of an item |
CN104199857A (en) * | 2014-08-14 | 2014-12-10 | 西安交通大学 | Tax document hierarchical classification method based on multi-tag classification |
CN105677838A (en) * | 2016-01-04 | 2016-06-15 | 中国科学院计算技术研究所 | User profile creating and personalized search ranking method and system based on user requirements |
CN106570197A (en) * | 2016-11-15 | 2017-04-19 | 北京百度网讯科技有限公司 | Searching and ordering method and device based on transfer learning |
CN106815252A (en) * | 2015-12-01 | 2017-06-09 | 阿里巴巴集团控股有限公司 | A kind of searching method and equipment |
-
2018
- 2018-11-27 CN CN201811422626.6A patent/CN109582868A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20110184806A1 (en) * | 2010-01-27 | 2011-07-28 | Ye Chen | Probabilistic recommendation of an item |
CN104199857A (en) * | 2014-08-14 | 2014-12-10 | 西安交通大学 | Tax document hierarchical classification method based on multi-tag classification |
CN106815252A (en) * | 2015-12-01 | 2017-06-09 | 阿里巴巴集团控股有限公司 | A kind of searching method and equipment |
CN105677838A (en) * | 2016-01-04 | 2016-06-15 | 中国科学院计算技术研究所 | User profile creating and personalized search ranking method and system based on user requirements |
CN106570197A (en) * | 2016-11-15 | 2017-04-19 | 北京百度网讯科技有限公司 | Searching and ordering method and device based on transfer learning |
Non-Patent Citations (2)
Title |
---|
WILFRED NG等: ""Mining User preference using Spy voting for search engine personalization"", 《ACM TRANSACTIONS ON INTERNET TECHNOLOGY》 * |
韩露: ""基于用户兴趣模型的个性化搜索系统的研究"", 《中国优秀博硕士学位论文全文数据库(硕士) 信息科技辑》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111274808A (en) * | 2020-02-11 | 2020-06-12 | 支付宝(杭州)信息技术有限公司 | Text retrieval method, model training method, text retrieval device, and storage medium |
EP4145308A4 (en) * | 2020-05-19 | 2023-09-20 | Huawei Technologies Co., Ltd. | Search recommendation model training method, and search result sorting method and device |
CN111782871A (en) * | 2020-06-18 | 2020-10-16 | 湖南大学 | Cross-modal video time positioning method based on space-time reinforcement learning |
CN111782871B (en) * | 2020-06-18 | 2024-06-25 | 湖南大学 | Cross-modal video time positioning method based on space-time reinforcement learning |
CN112487274A (en) * | 2020-12-02 | 2021-03-12 | 重庆邮电大学 | Search result recommendation method and system based on text click rate |
CN112487274B (en) * | 2020-12-02 | 2023-02-07 | 重庆邮电大学 | Search result recommendation method and system based on text click rate |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Chu et al. | A hybrid recommendation system considering visual information for predicting favorite restaurants | |
CN108573411B (en) | Mixed recommendation method based on deep emotion analysis and multi-source recommendation view fusion of user comments | |
CN107180045B (en) | Method for extracting geographic entity relation contained in internet text | |
CN106709040B (en) | Application search method and server | |
CN110442777B (en) | BERT-based pseudo-correlation feedback model information retrieval method and system | |
US7844592B2 (en) | Ontology-content-based filtering method for personalized newspapers | |
AU2011269676B2 (en) | Systems of computerized agents and user-directed semantic networking | |
CN111191122A (en) | Learning resource recommendation system based on user portrait | |
KR20190118477A (en) | Entity recommendation method and apparatus | |
CN105808590B (en) | Search engine implementation method, searching method and device | |
Nabli et al. | Efficient cloud service discovery approach based on LDA topic modeling | |
Chen et al. | Research on personalized recommendation hybrid algorithm for interactive experience equipment | |
WO2021196541A1 (en) | Method, apparatus and device used to search for content, and computer-readable storage medium | |
CN112307182B (en) | Question-answering system-based pseudo-correlation feedback extended query method | |
CN110888991A (en) | Sectional semantic annotation method in weak annotation environment | |
CN109582868A (en) | The search recommended method of preference is clicked based on term vector weighting, support vector regression and user | |
CN109816015B (en) | Recommendation method and system based on material data | |
CN111753167A (en) | Search processing method, search processing device, computer equipment and medium | |
CN111078835A (en) | Resume evaluation method and device, computer equipment and storage medium | |
Wei et al. | Online education recommendation model based on user behavior data analysis | |
Ma et al. | Content Feature Extraction-based Hybrid Recommendation for Mobile Application Services. | |
KR20130045425A (en) | Expert recommendation search method based on social ontology | |
KR101602342B1 (en) | Method and system for providing information conforming to the intention of natural language query | |
Zhang et al. | Personalized book recommender system based on Chinese library classification | |
CN111581326B (en) | Method for extracting answer information based on heterogeneous external knowledge source graph structure |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190405 |
|
WD01 | Invention patent application deemed withdrawn after publication |