CN110888990B

CN110888990B - Text recommendation method, device, equipment and medium

Info

Publication number: CN110888990B
Application number: CN201911179808.XA
Authority: CN
Inventors: 蔡远航; 郑少杰; 付勇; 范增虎
Original assignee: WeBank Co Ltd
Current assignee: WeBank Co Ltd
Priority date: 2019-11-22
Filing date: 2019-11-22
Publication date: 2024-04-12
Anticipated expiration: 2039-11-22
Also published as: CN110888990A; WO2021098648A1

Abstract

The invention discloses a text recommendation method, a device, equipment and a medium, and relates to the technical field of financial science and technology, wherein the method comprises the following steps: monitoring the operation behavior of the target user to determine keywords associated with the target user; retrieving more than one updated text containing at least one keyword from a preset text database set as a first candidate text; a preset event map is called, an updated text with the total association degree with the first candidate text not smaller than a preset association threshold value is selected from a preset text database set according to the map to serve as a second candidate text, and the event map comprises association relations between texts, wherein each association relation has the corresponding association degree; and screening selected texts from the first candidate texts and the second candidate texts according to the operation behaviors, and recommending the selected texts to the target user. The method and the device solve the technical problems of over-singleization of recommendation and low recommendation accuracy in the existing recommendation process.

Description

Text recommendation method, device, equipment and medium

Technical Field

The invention relates to the technical field of financial science (Fintech), in particular to a text recommendation method, a text recommendation device, text recommendation equipment and a text recommendation medium.

Background

Along with the development of computer technology, more and more technologies are applied in the financial field, the traditional financial industry is gradually changed to financial technology (Fintech), and content recommendation technology is not exceptional, but due to the requirements of safety, instantaneity and accuracy of the financial industry, higher requirements are also put forward on the content recommendation technology, at present, the content recommendation technology completely relies on keywords configured by users to recommend, and the keywords configured by the users are completely relied on to recommend the user preferences narrowly, so that the technical problems that news data and the like pushed to users are too single and have low accuracy exist.

Disclosure of Invention

The invention mainly aims to provide a text recommendation method, a device, equipment and a medium, and aims to solve the technical problems of too singleness of recommendation and low recommendation accuracy in the existing content recommendation process according to keywords.

In order to achieve the above object, an embodiment of the present invention provides a text recommendation method, including:

monitoring the operation behavior of a target user, and determining keywords associated with the target user according to the operation behavior;

retrieving more than one updated text containing at least one keyword from a preset text database set as a first candidate text;

A preset event map is called, an updated text with the total association degree with the first candidate text not smaller than a preset association threshold value is selected from the preset text database set according to the preset event map to serve as a second candidate text, and the event map comprises association relations between texts, wherein each association relation has the corresponding association degree;

and screening selected texts from the first candidate texts and the second candidate texts according to the operation behaviors, and recommending the selected texts to the target user.

Optionally, the step of retrieving a preset event map, and selecting, according to the preset event map, an updated text with a total association degree with the first candidate text not less than a preset association threshold from the preset text database set includes:

acquiring texts to be processed from the preset text database set at intervals of preset time periods;

html tag filtering, symbol filtering and clause processing are carried out on the text to be processed through a preset regular expression, so that a preprocessed text formed by a clause list is obtained;

and generating the preset rational map according to the preprocessing text.

Optionally, the step of generating the preset rational atlas according to the preprocessed text includes:

identifying a plurality of preset text association relations for each clause in the clause list to obtain a node text to be processed, wherein the preset text association relations comprise, but are not limited to, compliance, causality, conditions and parallel relations;

performing word segmentation on the node text to be processed through a preset word segmentation tool, acquiring word vectors of each word segment, and acquiring node vectors of each node text to be processed based on the word vectors of each word segment;

calculating a first node distance between each node text to be processed and other node texts to be processed according to the node vector of each node text to be processed;

performing iterative grafting treatment on two node texts to be treated, wherein the node distance of the node texts is smaller than a first preset distance, until each node text to be treated is in a convergence state that the node text relation edge is not changed any more, and setting each node text to be treated in the convergence state as a convergence node text;

and generating the preset rational map based on the convergent node text and the node text relation edge between the convergent node text.

Optionally, the step of retrieving a preset event map, selecting, according to the preset event map, an updated text with a total association degree with the first candidate text not less than a preset association threshold from the preset text database set, as the second candidate text includes:

a preset event map is called, and whether convergence node text containing the keywords in corresponding clauses exists in the event map is judged;

if yes, setting the convergence node text containing the keywords in the corresponding clause as a user attention node text, selecting a third candidate text outside the first candidate text from the texts updated in the preset time period in the preset text database, and performing word segmentation on the title of each text in the third candidate text through a preset word segmentation tool to obtain a title vector of each text in the third candidate text;

calculating a second node distance between the header vector and a node vector of each convergent node text in the rational atlas;

selecting a second node distance from the third candidate text to be smaller than a second preset distance, wherein the convergence node text corresponding to the second distance is a first target text of the user attention node text, or selecting a second node distance from the third candidate text to be smaller than the second preset distance, and the second target text of the user attention node text exists in a preset screening logic depth range of the convergence node text corresponding to the second distance, wherein the screening logic depth is determined according to the association degree of each association relation in the rational map;

And setting the first target text and the second target text as the second candidate text.

Optionally, the step of screening the selected text from the first candidate text and the second candidate text according to the operation behavior, and recommending the selected text to the target user includes:

acquiring the propagation quantity of each text in the first candidate text and the second candidate text, acquiring the correlation degree of each text and the target user, and acquiring the preference degree of the target user according to the operation behavior;

and screening selected texts from the first candidate texts and the second candidate texts according to the propagation quantity, the relevance and the preference, and recommending the selected texts to the target user.

Optionally, the step of obtaining the relevance between each text and the target user includes:

acquiring the number of times of occurrence of the keyword in each text of the first candidate text, and setting the number of times as word times;

the method comprises the steps of obtaining positions of keywords in each text of a first candidate text, setting the positions as word positions, and obtaining corresponding preset position weights of the word positions, wherein the word positions are different and the position weights are different, and the word positions comprise a first sentence position of a text first paragraph, a first sentence position of a text tail paragraph, a non-first sentence position of the text first paragraph, a non-first sentence position of the text tail paragraph, a non-first sentence position of the non-first paragraph and a non-second sentence position of the non-first sentence;

Acquiring the ratio of the number of sentences of the interval between the first and last occurrence positions of the keyword in each text of the first candidate text to the total number of sentences of the whole text, and setting the ratio as a word span;

obtaining a target text between the first and last occurrence positions of the keywords in each text of the first candidate text, obtaining the number of the keywords contained in each preset sentence number in the target text, and setting the number of the keywords contained in each preset sentence number as word density;

acquiring a first relevance of each text in the first candidate text according to the number of words, the position weight corresponding to the word position, the word span and the word density;

and obtaining screening logic depth of each text in the second candidate text, and determining a second relevance of each text in the second candidate text according to the screening logic depth.

Optionally, the step of obtaining the preference degree of the target user according to the operation behavior includes:

acquiring historical browsing texts of the target user from the operation behaviors, acquiring a first document vector of each text in the historical browsing texts, and acquiring a second document vector of each text in the first candidate text and the second candidate text;

And acquiring a first pearson correlation coefficient between the second document vector and the first document vector, and acquiring the preference degree of the target user according to the first pearson correlation coefficient.

Optionally, the step of obtaining the first document vector of each text in the historical browsing text includes:

acquiring a first probability matrix of each text in the historical browsing text, which is divided into a first preset category, according to a preset clustering algorithm;

obtaining word segmentation words of each text in the historical browsing text according to a preset word segmentation algorithm, and obtaining a second probability matrix of the word segmentation words divided into a second preset category;

acquiring each optimized word vector corresponding to each text in the historical browsing text according to the first probability matrix and the second probability matrix;

and acquiring a first document vector of each text in the historical browsing text according to the optimized word vector.

Optionally, the step of obtaining a first pearson correlation coefficient between the second document vector and the first document vector, and obtaining the preference degree of the target user according to the first pearson correlation coefficient includes:

acquiring historical browsing time from when each text in the historical browsing text is clicked to browse to the current moment;

Acquiring a first pearson correlation coefficient between the second document vector and the first document vector, and performing interest weight reduction processing on the first pearson correlation coefficient according to the historical browsing time to acquire a second pearson correlation coefficient;

and obtaining the preference degree of the target user according to the second pearson correlation coefficient.

Optionally, the step of screening the selected text from the first candidate text and the second candidate text according to the propagation amount, the relevance and the preference, and recommending the selected text to the target user includes:

calculating a value score of each text in the first candidate text and the second candidate text according to the propagation quantity, the first relevance, the second relevance and the preference;

and sequentially selecting a preset number of texts as selected texts according to the value score from high to low, and recommending the selected texts to the target user.

The invention also provides a text recommendation device, which comprises:

the monitoring module is used for monitoring the operation behaviors of the target user and determining keywords associated with the target user according to the operation behaviors;

The searching module is used for searching more than one updated text containing at least one keyword from a preset text database set and taking the updated text as a first candidate text;

the selection module is used for retrieving a preset event map, selecting updated texts with the total association degree with the first candidate texts not smaller than a preset association threshold value from the preset text database set according to the preset event map, and taking the updated texts as second candidate texts, wherein the event map comprises association relations between texts, and each association relation has the corresponding association degree;

and the screening module is used for screening the selected text from the first candidate text and the second candidate text according to the operation behaviors and recommending the selected text to the target user.

Optionally, the text recommendation device further includes:

the acquisition module is used for acquiring texts to be processed from the preset text database set every interval preset time period;

the preprocessing module is used for carrying out html tag filtering, symbol filtering and clause processing on the text to be processed through a preset regular expression to obtain a preprocessed text formed by a clause list;

And the generation module is used for generating the preset rational map according to the preprocessing text.

Optionally, the generating module includes:

the recognition unit is used for recognizing a plurality of preset text association relations for each clause in the clause list to obtain a node text to be processed, wherein the preset text association relations comprise, but are not limited to, compliance, cause and effect, conditions and parallel relations;

the first acquisition unit is used for carrying out word segmentation on the node text to be processed through a preset word segmentation tool, acquiring word vectors of each word segment, and acquiring the node vector of each node text to be processed based on the word vector of each word segment;

the first calculation unit is used for calculating a first node distance between each node text to be processed and other node texts to be processed according to the node vector of each node text to be processed;

the grafting processing unit is used for carrying out iterative grafting processing on two node texts to be processed, wherein the node distance of the node texts to be processed is smaller than the first preset distance, until each node text to be processed is in a convergence state that the node text relation edge is not changed any more, and each node text to be processed in the convergence state is set as a convergence node text;

And the generation unit is used for generating the preset event map based on the convergent node text and the node text relation edge between the convergent node texts.

Optionally, the selecting module includes:

the invoking unit is used for invoking a preset event map and judging whether a convergent node text containing the keyword in a corresponding clause exists in the event map;

the first setting unit is used for setting the convergence node text containing the keywords in the corresponding clause as a user attention node text, selecting a third candidate text outside the first candidate text from the texts updated in the preset time period in the preset text database, and performing word segmentation on the title of each text in the third candidate text through a preset word segmentation tool to obtain a title vector of each text in the third candidate text;

a second calculating unit, configured to calculate a second node distance between the header vector and a node vector of each convergent node text in the rational map;

a selecting unit, configured to select, from the third candidate texts, a second node distance smaller than a second preset distance, where the convergence node text corresponding to the second preset distance is a first target text of the node text concerned by the user, or select, from the third candidate texts, a second node distance smaller than the second preset distance, where a second target text of the node text concerned by the user exists in a preset screening logic depth range of the convergence node text corresponding to the second preset distance, where the screening logic depth is determined according to a degree of association of each association relationship in the event map;

And the second setting unit is used for setting the first target text and the second target text as the second candidate text.

Optionally, the screening module includes:

the second acquisition unit is used for acquiring the propagation quantity of each text in the first candidate text and the second candidate text, acquiring the correlation degree of each text and the target user, and acquiring the preference degree of the target user according to the operation behavior;

and the recommending unit is used for screening selected texts from the first candidate texts and the second candidate texts according to the propagation quantity, the relevance and the preference, and recommending the selected texts to the target user.

Optionally, the second acquisition unit includes:

a first obtaining subunit, configured to obtain a number of times that the keyword appears in each text of the first candidate text, and set the number of times as a word number;

the second obtaining subunit is configured to obtain a position of the keyword in each text of the first candidate text, set the position as a word position, and obtain a position weight corresponding to the word position, where the word position is different and the position weight is different, where the word position includes a first text paragraph position, a last text paragraph position, a first text paragraph non-first sentence position, a last text non-first sentence position, a non-first paragraph first sentence position, and a non-last paragraph first sentence position;

A third obtaining subunit, configured to obtain a ratio of the number of sentences in the interval between the positions of the first and last occurrence of the keyword in each text of the first candidate text to the total number of sentences in the whole text, and set the ratio as a word span;

a fourth obtaining subunit, configured to obtain a target text between positions where the keyword appears for the first time and the last time in each text of the first candidate text, obtain an average number of keywords included in each preset number of sentences in the target text, and set the average number of keywords included in each preset number of sentences as word density;

a fifth obtaining subunit, configured to obtain a first relevance of each text in the first candidate text according to the number of times of the word, a position weight corresponding to the word position, the word span, and the word density;

and a sixth obtaining subunit, configured to obtain a screening logic depth of each text in the second candidate text, and determine a second relevance of each text in the second candidate text according to the screening logic depth.

Optionally, the second acquisition unit includes:

a seventh obtaining subunit, configured to obtain, from the operation behavior, a history browsing text of the target user, obtain a first document vector of each text in the history browsing text, and obtain a second document vector of each text in the first candidate text and the second candidate text;

An eighth obtaining subunit, configured to obtain a first pearson correlation coefficient between the second document vector and the first document vector, and obtain the preference degree of the target user according to the first pearson correlation coefficient.

Optionally, the seventh acquisition subunit is configured to implement:

Optionally, the eighth obtaining subunit is configured to implement:

Optionally, the screening module includes:

a third calculation unit, configured to calculate a value score of each text in the first candidate text and the second candidate text according to the propagation amount, the first relevance, the second relevance, and the preference;

and the screening unit is used for sequentially selecting a preset number of texts from high to low according to the value score to serve as selected texts, and recommending the selected texts to the target user.

The invention also provides a medium, wherein the medium is stored with a text recommendation program, and the text recommendation program realizes the steps of the text recommendation method when being executed by a processor.

The method comprises the steps of monitoring the operation behaviors of a target user, and determining keywords associated with the target user according to the operation behaviors; after obtaining keywords, retrieving more than one updated text containing at least one keyword from a preset text database set as a first candidate text; after a first candidate text is obtained, a preset event map is called, an updated text with the total association degree with the first candidate text not smaller than a preset association threshold value is selected from the preset text database set according to the preset event map to serve as a second candidate text, the selection category of the candidate text in the recommendation process is enlarged by obtaining the second candidate text, and the event map comprises association relations between texts, wherein the association relations have the corresponding association degrees; and screening selected texts from the first candidate texts and the second candidate texts according to the operation behaviors, and recommending the selected texts to the target user. In the application, the selected text is not selected singly from the first candidate texts searched according to the keywords, but selected from the second candidate texts and the first candidate text set obtained according to the preset event map and the like, so that the content recommendation is prevented from being singulated, and the recommendation accuracy can be improved because the content is recommended according to the association relation between the comprehensive reference text and the text instead of the keyword alone.

Drawings

FIG. 1 is a flowchart of a text recommendation method according to a first embodiment of the present invention;

FIG. 2 is a schematic diagram of a refinement flow before a text updating step of selecting a text update with a total association degree with the first candidate text not less than a preset association threshold from the preset text database set according to a preset event map based on a calling of the preset event map in a second embodiment of the text recommendation method of the present invention;

FIG. 3 is a schematic diagram of a device architecture of a hardware operating environment involved in a method according to an embodiment of the present invention;

FIG. 4 is a schematic view of a first scenario involved in the text recommendation method of the present invention;

FIG. 5 is a schematic diagram of a second scenario involved in the text recommendation method of the present invention;

FIG. 6 is a schematic diagram of a third scenario involved in the text recommendation method of the present invention;

FIG. 7 is a schematic diagram of a fourth scenario involving a text recommendation method of the present invention;

FIG. 8 is a schematic diagram of a fifth scenario involving a text recommendation method of the present invention;

FIG. 9 is a schematic diagram of a sixth scenario involved in the text recommendation method of the present invention;

FIG. 10 is a schematic diagram of a seventh scenario involving a text recommendation method of the present invention;

FIG. 11 is a schematic view of an eighth scenario involved in the text recommendation method of the present invention;

FIG. 12 is a schematic illustration of a ninth scenario involving a text recommendation method of the present invention;

Fig. 13 is a schematic flow chart of the text recommendation method according to the present invention.

The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.

Detailed Description

It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

The invention provides a text recommendation method, in an embodiment of the text recommendation method, referring to fig. 1, the text recommendation method includes:

step S10, monitoring the operation behavior of a target user, and determining keywords associated with the target user according to the operation behavior;

step S20, more than one updated text containing at least one keyword is searched from a preset text database set and used as a first candidate text;

step S30, a preset event map is called, updated texts with the total association degree with the first candidate texts not smaller than a preset association threshold value are selected from the preset text database set according to the preset event map to serve as second candidate texts, and the event map comprises association relations between texts, wherein the association relations have the corresponding association degrees;

And S40, screening out selected texts from the first candidate texts and the second candidate texts according to the operation behaviors, and recommending the selected texts to the target user.

The method comprises the following specific steps:

at present, more and more public opinion systems appear on the market to meet the demands of enterprises on monitoring network public opinion and individual topic tracking of hot events, and specifically, the public opinion systems can help the enterprises to realize functions of listening to target audience ideas, analyzing industry trends, managing brand reputation, performing crisis early warning and the like, and the current public opinion systems generally realize the functions through the following processes: 1. and (3) data acquisition: collecting all information sources of the whole network, wherein all information sources of the whole network comprise news web media, forums, blogs, microblogs, various information clients and the like; 2. data screening: screening news data according to monitoring task keywords configured on a public opinion system, for example, if the text of a certain news contains keywords configured by a user, reserving the certain news for subsequent processing; 3. and (3) data processing: for all news containing keywords, sequentially calculating emotion tendencies of texts, propagation quantity of news, correlation degree between news texts and the keywords and the like; 4. data pushing: the emotion, propagation quantity, correlation degree and click preference of a user on a public opinion system for historically pushing news data are comprehensively considered, the news data processed in the steps 1-3 are sorted, and a plurality of news which are most likely to be interested by the user are selected for pushing, namely in the prior art, recommendation of target content is carried out completely depending on keywords configured by the user and the click preference of the user for the historically pushing news data, in the process of pushing the target content, word2vec (a word vector model which can describe semantic similarity of Chinese words according to the distance between word vectors) word vectors are used for processing when the user preference is analyzed, and word2vec word vectors are used for processing when the user preference is analyzed, so that the preference of the user is narrowly defined, and the pushed content is single. Specifically, after the user clicks a news item such as "enterprise a cooperates with bank B" for a plurality of times in a certain period of time and word2vec word vector processing is performed, the public opinion system learns that the user prefers news between enterprise a and bank B. If another news appears that "enterprise a and bank C together build a laboratory in hundreds of millions at a certain university investment", the public opinion system will not consider the news that is preferred or liked by the user with a high probability, and the actual preference of the user is the layout situation of enterprise a in the financial field, while pushing only the news between enterprise a and bank B obviously results in the singulation of the pushed content. In addition, in the target content pushing process, the recommendation of the target content is performed completely depending on the keyword configured by the user and the click preference of the user on the history pushing news data, so that the technical problems of too single recommendation and low recommendation accuracy of the content such as the news data recommended to the user can be caused. Therefore, the data screening completely depends on the keywords, and the recommendation accuracy is low in various scenes.

In order to solve the above technical problem, in this embodiment, the operation behavior of the target user is monitored, the keyword associated with the target user is determined according to the operation behavior, where the operation behavior includes a sliding behavior or a search behavior triggered by inputting a keyword, and if the operation behavior is a sliding behavior, the determination of the keyword associated with the target user according to the operation behavior may be: extracting a pre-stored keyword associated with the target user, wherein the pre-stored keyword associated with the target user is obtained based on a historical browsing text of the target user, and if the operation behavior is a search behavior of an input keyword, the keyword associated with the target user may be the input keyword, or may be a combination of the input keyword and the pre-stored keyword associated with the target user.

the preset text database set comprises a database set formed by news web media, forums, blogs, microblogs, other various information clients and the like.

Retrieving, from a preset text database set, one or more updated texts including at least one keyword as a first candidate text, and retrieving, from the preset text database set, one or more updated texts including at least one keyword as a first candidate text, where the first candidate text may be: retrieving, in real time, one or more updated texts containing at least one keyword from a preset text database set as a first candidate text (for facilitating real-time recommendation), or retrieving, in real time, one or more updated texts containing at least one keyword from the preset text database set at intervals of a certain period of time as a first candidate text (for facilitating timing recommendation), or retrieving one or more updated texts containing at least one keyword from the preset text database set at this time only as a first candidate text (for facilitating recommendation when searching by a target user), and the like.

the preset rational map is generated and updated in real time or at regular time, the total association degree and the like can be associated with logic depth between words or distance between words and the like, and the association relationship between texts can be causal, compliant and the like, and the association degrees of the causal, compliant and the like are different. And calling a preset event map, and selecting an updated text with the total association degree with the first candidate text not smaller than a preset association threshold value from the preset text database set according to the preset event map as a second candidate text, wherein the recommendation range of the second candidate text is improved in the content recommendation process.

As shown in fig. 2, the step of retrieving a preset rational map, and selecting, according to the preset rational map, an updated text with a total association degree with the first candidate text not less than a preset association threshold from the preset text database set includes:

A1, acquiring a text to be processed from the preset text database set every preset time period;

a2, carrying out html tag filtering, symbol filtering and clause processing on the text to be processed through a preset regular expression to obtain a preprocessed text formed by a clause list;

and A3, generating the preset rational map according to the preprocessing text.

In this embodiment, the text to be processed is collected from the preset text database set at each preset time interval (may include real time), where, since the amount of text to be processed collected every day is in the tens of millions, a preset collection model, such as a preset spark streaming model, may be used to complete the collection. After obtaining a text to be processed, performing html tag filtering, symbol filtering and clause processing on the text to be processed through a preset regular expression to obtain a preprocessed text formed by a clause list, optionally, collecting the text to be processed from the preset text database set every preset time period, after obtaining the text to be processed, performing html tag filtering, symbol filtering and clause processing on each text in the text to be processed through the preset regular expression to obtain the preprocessed text formed by the clause list, specifically, filtering out the html tag in each text body of the text to be processed through the following 4 regular expressions, wherein the first rule is: '// < ++cdata [ [ ]// [ ] >, second bar: 's' script [ ] ] > [ </s/s 'script\s', third bar: 'style [ ++ > ] [ < ] </s/s style \s >', fourth bar: ' </i > - [ - > ], filtering html tags in a text by the 4 regular expressions, filtering out emoticons in each text of the text to be processed by the following 4 regular expressions, and filtering out the first text: "\U0001F 600\U 0001F64F", second: "\U0001F 300\U 0001F5FF", third: "\U0001F 680\U 0001F6FF", fourth: "\U0001F1E 0\U 0001F1FF", after filtering, the text to be processed, such as the body of each text, is subjected to clauses: according to punctuation marks. ","? ","! The text is split into a list of sentences by "," | ", etc., in this embodiment, although the sentence processing is separately performed for each text, the sentences of each text in the list of sentences may be mixed instead of being discriminated according to each text.

The step of generating the preset rational map according to the preprocessing text comprises the following steps:

step A31, identifying a plurality of preset text association relations for each clause in the clause list to obtain a node text to be processed, wherein the preset text association relations comprise, but are not limited to, compliance, cause and effect, conditions and parallel relations;

and identifying a plurality of preset text association relations for each clause in the clause list, wherein the preset text association relations comprise types such as compliance, causality, conditions, parallel relations and the like, and the node text to be processed is obtained after identifying the plurality of preset text association relations for each clause in the clause list.

Specifically, two events (two event phrases) of the table compliance/causal/condition/juxtaposition relationship can be identified from each text sentence of each text of the pre-processed text according to the preset event word segmentation tool, the two event phrases can be set as the node text to be processed in the event map, and the event map connects the two node text to be processed by using one directed edge (directed line), for example, the node text to be processed as shown in fig. 4 can be obtained from a text sentence of which the line drop information makes the loan cost low, and in this embodiment, the table compliance/causal/condition/juxtaposition relationship node text to be processed can be identified from each text sentence of each text of the pre-processed text based on a preset text association relation extraction model, where the preset table compliance/causal/condition/juxtaposition relationship word combination in the preset text association relation includes: (first, second), (first, then), (first, then) and so on, if one sentence contains two words in a phrase of the word combination carried by the table at the same time, and the appearance sequence of the two words in the sentence is consistent with the sequence defined in the phrase, extracting the two clauses led by the two words through a preset leading clause model in a preset text association relation extraction model, removing all punctuation marks (preset), mood words (preset), auxiliary words (preset), stop words (preset) and so on in the clause as two node texts to be processed in a case map, and simultaneously connecting the two node texts to be processed by using a directed edge (edge of a logical relation) carried by the table. For example, the sentence "first a, second B", is processed as shown in fig. 5.

Likewise, the word combinations of the preset causal agents are: (because, so), (because, cause), (because, so), (since, then), (since, once, (due to, therefore), (due to, cause), (due to, thus), (_, resulting), (_, thus), etc. The underline "_" in the phrase indicates the blank word, and the blank word can be ignored for matching in the subsequent matching process. If a sentence contains two words in a phrase of the word combination of the table cause and effect, the appearance sequence of the two words in the sentence is consistent with the sequence defined in the phrase, the two clauses guided by the two words are extracted through a preset guiding clause model, all punctuation marks (preset), intonation words (preset), auxiliary words (preset), stop words (preset) and the like in the clauses are removed as two node texts to be processed in the atlas, and the two node texts to be processed are connected by a directed edge (edge of a logic relation) of the table cause and effect. For example, sentence "because of a, B", processing results in fig. 6.

Likewise, the word combinations of the preset table conditions are: (if so), (if so), (once), (as long as it is), (if so), (only) and so on. If a sentence contains two words in a phrase in the word combination of the table conditions, and the appearance sequence of the two words in the sentence is consistent with the sequence defined in the phrase, the two phrases guided by the two words are extracted through a preset guiding clause model, all punctuation marks (preset), intonation words (preset), auxiliary words (preset), stop words (preset) and the like in the phrases are removed as two node texts to be processed in the atlas, and the two node texts to be processed are connected by using the directed edges (edges of the logic relationship) of one table condition, for example, the sentence "if A, B" is obtained after processing, and fig. 7 is obtained.

Similarly, the preset list of parallel words is combined with: (not only, but also), (not only, and), (not only, but also), (not only, and), (not only, and, (not only), (but also), (not only, and), (not only, but also), (either, or), (or), etc. If a sentence contains two words in a phrase of the word combination in parallel with the table, and the appearance sequence of the two words in the sentence is consistent with the sequence defined in the phrase, extracting two clauses guided by the two words through a preset guiding clause model, removing all punctuation marks (preset), intonation words (preset), auxiliary words (preset), stop words (preset) and the like in the clauses to serve as two node texts to be processed in the atlas, and simultaneously connecting the two node texts to be processed by using a directed edge (edge of a logic relation) in parallel with the table. For example, the sentence "not only a but also B", and fig. 8 is obtained after processing.

It should be noted that, if the preprocessed text includes text data greater than a preset number, such as hundred thousand, the event phrases of the plurality of preset text association relations in each text sentence in the preprocessed text may be extracted through a preset bidirectional extraction network model.

Step A32, word segmentation is carried out on the node text to be processed through a preset word segmentation tool, word vectors of each word segment are obtained, and the node vector of each node text to be processed is obtained based on the word vector of each word segment;

the method comprises the steps of performing word segmentation on a node text to be processed through a preset word segmentation tool such as a preset barking word segmentation tool (an open-source Chinese word segmentation tool can be used for segmenting words and marking parts of speech of an input Chinese text), mapping each Chinese word into a high-dimensional vector (a 200-dimensional vector can be taken) through a preset word2vec (a word vector model) after word segmentation, and obtaining node vectors of each node text to be processed based on the word vectors of each word to be segmented by supposing that 5 high-dimensional vectors are obtained, abcde, and adding the word vectors of the 5 word vectors according to corresponding dimensions and dimension weights to obtain the word vector of each word to be segmented.

Step A33, calculating a first node distance between each node text to be processed and other node texts to be processed according to the node vector of each node text to be processed;

for any two Chinese words, the closer the semantics are, the closer the distance between the mapped vectors is, so that the semantic similarity of the Chinese words can be described according to the distance between word vectors.

In this embodiment, a first node distance between each node text to be processed and other node texts to be processed is calculated according to the node vector of each node text to be processed, specifically, a calculation formula of a pearson correlation coefficient of a preset node text is obtained, and a node text between two node text vectors to be processed is calculated according to the node vector of each node text to be processed and the calculation formula of the pearson correlation coefficient of the preset node textPearson correlation coefficient usingRepresenting the pearson correlation coefficient of the node text between two node text vectors to be processed, then the first node distance between the two node text vectors to be processed may be represented as +.>And for each node text to be processed, sequentially calculating the distance between the node text to be processed and all the rest node texts to be processed.

Step A34, performing iterative grafting treatment on two node texts to be treated, wherein the node distance of the node texts to be treated is smaller than a first preset distance, until each node text to be treated is in a convergence state that the node text relation edge is not changed any more, and setting each node text to be treated in the convergence state as a convergence node text;

and step A35, generating the preset event map based on the convergent node text and the node text relation edge between the convergent node texts.

Performing iterative grafting processing on two node texts to be processed, wherein the node distance is smaller than a first preset distance, until each node text to be processed is in a convergence state that node text relationship edges are not changed any more, wherein each node text to be processed in the convergence state is set to be a convergence node text, specifically, for example, if the distance between the node text to be processed A and a certain node B is found to be smaller than the first preset distance, for example, smaller than 0.3, all the relationships between the node text to be processed A are grafted to the node text to be processed B, and meanwhile, the node text to be processed A is deleted, as shown in fig. 9, if the distance between the node text to be processed A and the node text to be processed C is smaller than the first preset distance, for example, smaller than 0.3, the relationships between the node texts to be processed are obtained as shown in fig. 10, grafting treatment is carried out on two to-be-treated node texts with the node distance smaller than the first preset distance until each to-be-treated node text of the to-be-treated texts is in a convergence state, namely, the calculation process of carrying out grafting treatment on two to-be-treated node texts with the node distance smaller than the first preset distance is carried out iteratively until each to-be-treated node text of the to-be-treated texts is in a convergence state, so that the relation edges (relation boundaries) of the to-be-treated node texts forming a directed edge map are not changed any more, a to-be-treated map of the directed edge is generated, namely, the map is considered to have reached the convergence state, and it is required to treat the relation edge in the to-be-treated map in a fusion process, for example, the existence of a following node D of the directed edge in the table is assumed, the following node D of the directed edge is grafted on the father to-be-treated node text A of the leading node B of the directed edge, a same directed edge as between a and B was created to connect a and D, resulting in fig. 12.

And comprehensively screening selected texts from the first candidate text and the second candidate text after the second candidate text and the first candidate text are obtained, recommending the selected texts to the target user, and recommending the selected texts to the target user instead of only screening the selected texts from the first candidate text.

Further, based on the first embodiment, in another embodiment of the text recommendation method provided by the present invention, the step of retrieving a preset rational map, and selecting, as the second candidate text, an updated text having a total association degree with the first candidate text not less than a preset association threshold from the preset text database set according to the preset rational map includes:

step S31, a preset event map is called, and whether convergence node text containing the keywords in corresponding clauses exists in the event map is judged;

step S32, if yes, setting the convergence node text containing the keywords in the corresponding clause as a user attention node text, selecting a third candidate text outside the first candidate text from the texts updated in the preset time period in the preset text database, and performing word segmentation on the title of each text in the third candidate text through a preset word segmentation tool to obtain a title vector of each text in the third candidate text;

and calling a preset event map, judging whether a convergence node text containing the keyword in a corresponding sentence exists in the event map, wherein the keyword can be contained in the corresponding sentence, or the keyword can not be contained in the corresponding sentence, if the convergence node text containing the keyword in the corresponding sentence does not exist in the event map, carrying out subsequent processing, directly selecting a selected text from a first candidate text to recommend, if the convergence node text containing the keyword in the corresponding sentence exists in the event map, marking the node text to be processed as a user attention node text, selecting a third candidate text outside the first candidate text from the text updated in a preset time period in the preset text database, carrying out word segmentation processing on the title of each text in the third candidate text through a preset word segmentation tool to obtain a title vector of each text in the third candidate text, specifically, carrying out preset word segmentation on the title of each text in the third candidate text, and obtaining a title vector of each text in the third candidate text by means of a preset word2 word, and calculating the title vector of each text in the third candidate text by means of a preset word to obtain a second title vector of each text in the third candidate text.

Step S33, calculating a second node distance between the title vector and the node vector of each convergent node text in the event map;

step S34, selecting a second node distance smaller than a second preset distance from the third candidate texts, wherein the convergence node text corresponding to the second distance smaller than the second preset distance is a first target text of the node text concerned by the user, or selecting a second target text with the second node distance smaller than the second preset distance from the third candidate texts, and the second target text of the node text concerned by the user exists in a preset screening logic depth range of the convergence node text corresponding to the second distance smaller than the second preset distance, wherein the screening logic depth is determined according to the association degree of each association relation in the event map;

calculating a second node distance between the header vector and the node vector of each converging node text in the event map, if the distance is smaller than a second preset distance, such as smaller than 0.4, and the node text to be processed is a "user attention node text", determining that the header vector corresponds to the text as a first target text, and if the distance is smaller than the second preset distance, such as smaller than 0.4, and other nodes marked as "user attention node text" exist in a range of a preset screening logic depth, such as a logic depth of 2, of the node text to be processed, retaining the text corresponding to the header vector as a second target text, wherein the screening logic depth is determined according to the association degree of each association relation in the event map, namely, the screening logic depth can be defined as follows: the logic depth of the edge of the table 'parallel' logic relationship is recorded as 0.5, the logic depth of the edge of the table 'parallel' logic relationship is recorded as 0.7, the logic depth of the edge of the table 'causal' logic relationship and the table 'conditional' logic relationship is recorded as 1, the logic depth between two node texts to be processed is the sum of the logic depths of all edges between the nodes, for example, the association between the node text to be processed B and the node text C concerned by a user can be realized through the two edges at the highest, and the two edges respectively represent the causal 'logic relationship and the parallel' logic relationship, the screening logic depth of the node text to be processed B or the logic depth of the text corresponding to a title vector is 1.7, and the 1.7 is in the logic depth of 2.

And step S35, setting the first target text and the second target text as the second candidate text.

And after the first target text and the second target text are obtained, setting the first target text and the second target text as the second candidate text.

In this embodiment, by calling a preset event map, whether a convergence node text containing the keyword in a corresponding clause exists in the event map is determined; if yes, setting the convergence node text containing the keywords in the corresponding clause as a user attention node text, selecting a third candidate text outside the first candidate text from the texts updated in the preset time period in the preset text database, and performing word segmentation on the title of each text in the third candidate text through a preset word segmentation tool to obtain a title vector of each text in the third candidate text; calculating a second node distance between the header vector and a node vector of each convergent node text in the rational atlas; selecting a second node distance from the third candidate text to be smaller than a second preset distance, wherein the convergence node text corresponding to the second distance is a first target text of the user attention node text, or selecting a second node distance from the third candidate text to be smaller than the second preset distance, and the second target text of the user attention node text exists in a preset screening logic depth range of the convergence node text corresponding to the second distance, wherein the screening logic depth is determined according to the association degree of each association relation in the rational map; and setting the first target text and the second target text as the second candidate text. The embodiment realizes accurate acquisition of the second candidate text and lays a foundation for accurate text recommendation.

Further, on the basis of the foregoing embodiment, in another embodiment of the present invention, the step of screening the selected text from the first candidate text and the second candidate text according to the operation behavior, and recommending the selected text to the target user includes:

step S41, acquiring the propagation amount of each text in the first candidate text and the second candidate text, acquiring the correlation degree of each text and the target user, and acquiring the preference degree of the target user according to the operation behavior;

searching the title of each text in the first candidate text and the second candidate text in a preset search engine to obtain the propagation quantity of each text, wherein the propagation quantity of each text reflects the heat of the text, and in the embodiment, the following is considered: two news having the same title belong to two forwarding of the same news, and the steps of calculating the propagation amount can be as follows: firstly deleting all punctuations in the titles of the first candidate text and the second candidate text (because some punctuations may be modified from half angle to full angle in the text acquisition process, in addition, some media also modify partial punctuations from half angle to full angle or from full angle to half angle in the text forwarding process, so when calculating the propagation quantity, the difference of the punctuations in the titles is not considered), then retrieving a preset number of texts such as 1000 texts from a preset search engine by using the title with all the punctuations deleted (in general, the maximum forwarding quantity of one text is in the order of hundreds and not more than 1000), deleting the retrieved titles of 1000 texts in sequence, and counting and comparing the current first candidate text and the current punctuations The number of the titles of the second candidate text is completely consistent as the propagation quantity of the current news。

The step of obtaining the relevance between each text and the target user comprises the following steps:

step S41, obtaining the number of times of occurrence of the keyword in each text of the first candidate text, and setting the number of times as word number;

step S42, obtaining the position of the keyword in each text of the first candidate text, setting the position as a word position, and obtaining a corresponding preset position weight of the word position, wherein the word positions are different and the position weights are different, and the word positions comprise a first text paragraph position, a last text paragraph position, a first text paragraph non-first sentence position, a last text paragraph non-first sentence position, a non-first paragraph first sentence position and a non-last paragraph first sentence position;

step S43, obtaining the ratio of the number of sentences of the interval between the first and last occurrence positions of the keyword in each text of the first candidate text to the total number of sentences of the whole text, and setting the ratio as word span;

step S44, obtaining a target text between the first and last occurrence positions of the keywords in each text of the first candidate text, obtaining the number of keywords contained in each preset sentence number in the target text, and setting the number of keywords contained in each preset sentence number as word density;

Step S45, obtaining a first relevance of each text in the first candidate text according to the number of times of words, the corresponding preset position weight of the word position, the word span and the word density;

the relevance calculations of the first candidate text and the second candidate text are different, as shown in fig. 13.

The first relevance of the first candidate text is calculated as follows: acquiring word times, word positions and word spans of the keywords in each text of the first candidate textAnd the density and the like, and acquiring a first relevance of each text of the first candidate text according to the word times, the word positions and the word spans, wherein the word times a: the total number of occurrences of keywords in the text body; word position b: b is 0 at the initial time, if the keyword appears in the first sentence of the first segment or the first sentence of the last segment of the text body, adding 2 to b; if the keyword appears in the first segment non-first sentence or the last segment non-first sentence of the text body, adding 1 to b; if the keyword appears in the first sentence of the rest paragraphs except the first paragraph and the tail paragraph, adding 0.5 to b; word span c: the ratio of the number of sentences of the keyword spaced between the first and last occurrence positions in the text body to the total number of sentences of the text; word density d: intercepting the text between the first and last positions of the keywords in the text, wherein in the text, the number of keywords contained in every preset level sentence, such as every 10 sentences, is defined as word density d, and then the relevance calculation formula is as follows: 。

Step S46, obtaining screening logic depth of each text in the second candidate texts, and determining second relevance of each text in the second candidate texts according to the screening logic depth.

The second candidate text does not contain keywords, and thus, a second relevance of each text of the second candidate text is determined according to the filtering logic depth, specifically, the second relevance is defined asWherein->The filtering logic depth is represented, and specifically, the filtering logic depth refers to the logic depth existing between the initial node text of the event map and the corresponding user attention node text in the process of filtering the second candidate text or the depth of the least logic edge contained in the logic depth.

The step of obtaining the preference degree of the target user according to the operation behavior comprises the following steps:

step S47, acquiring a historical browsing text of the target user from the operation behaviors, acquiring a first document vector of each text in the historical browsing text, and acquiring a second document vector of each text in the first candidate text and the second candidate text;

step S48, a first pearson correlation coefficient between the second document vector and the first document vector is obtained, and the preference degree of the target user is obtained according to the first pearson correlation coefficient.

In this embodiment, a history browsing text of the target user text is obtained, where the history browsing text may be a history browsing text in the past month, a first document vector of each text in the history browsing text is obtained, a second document vector of each text in the first candidate text and the second candidate text is obtained, a first pearson correlation coefficient between the second document vector and the first document vector is obtained, a preference degree of the target user is obtained according to the first pearson correlation coefficient, and a user preference degree is obtained according to the history browsing text, the first candidate text and the second candidate text.

And step S42, screening selected texts from the first candidate texts and the second candidate texts according to the propagation quantity, the relevance and the preference, and recommending the selected texts to the target user.

In this embodiment, the propagation amount, the relevance and the preference are integrated, a selected text is selected from the first candidate text and the second candidate text, and the selected text is recommended to the target user.

In this embodiment, the preference degree of the target user is obtained according to the operation behavior by obtaining the propagation amount of each text in the first candidate text and the second candidate text and obtaining the correlation degree of each text and the target user; and screening selected texts from the first candidate texts and the second candidate texts according to the propagation quantity, the correlation degree and the preference degree, and recommending the selected texts to the target user.

Further, on the basis of the foregoing embodiment, in another embodiment of the present invention, a text recommendation method is provided, where the step of obtaining the first document vector of each text in the history browsing text includes:

step B1, acquiring a first probability matrix of each text in the historical browsing text, which is divided into a first preset category, according to a preset clustering algorithm;

step B2, word segmentation words of each text in the historical browsing text are obtained according to a preset word segmentation algorithm, and a second probability matrix of the word segmentation words divided into a second preset category is obtained;

step B3, obtaining each optimized word vector corresponding to each text in the historical browsing text according to the first probability matrix and the second probability matrix;

and B4, acquiring a first document vector of each text in the historical browsing text according to the optimized word vector.

According to a preset clustering algorithm, a first probability matrix of each text in the historical browsing text divided under a first preset category (comprising 200 text subcategories) is obtained, specifically, an unsupervised clustering algorithm can be used for carrying out unsupervised clustering on the historical browsing text (the clustering number can be set to 200) by using an LDA (Latent Dirichlet Allocation) algorithm, so as to obtain a first probability matrix p of each text divided under the first preset category, according to a preset word segmentation algorithm, word segmentation words of each text in the historical browsing text are obtained, a second probability matrix q of the word segmentation words divided under a second preset category (comprising 200 word subcategories) is obtained, according to the first probability matrix and the second probability matrix, each optimized word vector W corresponding to each text in the historical browsing text is obtained, and W=0.6p+0.4q is obtained according to the optimized word vectors. Specifically, the optimized word vectors of all the segmented words are added to obtain the document vector of the corresponding news.

Acquiring a first pearson correlation coefficient between the second document vector and the first document vector, acquiring the preference degree of the target user according to the first pearson correlation coefficient,

acquiring a first pearson correlation coefficient between the second document vector and the first document vector, acquiring the preference degree of the target user according to the first pearson correlation coefficient, firstly searching a text set which is clicked and browsed by the current target user in history from a preset text database setA total of k texts. Then, respectively carrying out preset bargaining word segmentation processing on the first candidate text and the second candidate text and the k searched texts such as news, and adding word vectors of all the word segments to obtain a document vector of a corresponding text +.>. Use->Representing a history of browsed text sets +.>Document vector of each text in (a) with +.>Document vectors representing each of a current first candidate text and said second candidate text with +.>Representing two document vectors +.>A first pearson correlation coefficient therebetween, then the user preference of the first candidate text to the second candidate text may be expressed as

Wherein in the formula Document vector representing each of a first candidate text and said second candidate text,/->A document vector representing each text in the set of historically browsed text.

Wherein the step of obtaining the first pearson correlation coefficient between the second document vector and the first document vector, and obtaining the preference degree of the target user according to the first pearson correlation coefficient includes:

step C1, acquiring historical browsing time from when each text in the historical browsing text is clicked to browse to the current moment;

step C2, obtaining a first pearson correlation coefficient between the second document vector and the first document vector, and performing interest weight reduction processing on the first pearson correlation coefficient according to the historical browsing time to obtain a second pearson correlation coefficient;

and C3, acquiring the preference degree of the target user according to the second pearson correlation coefficient.

It is considered that the preferences of the target user may deviate significantly over time. For example, the text that is of greatest concern to an operator 1 week ago is a company's new product release meeting, while the text is currently of greatest concern is the public's assessment of a company's new product, etc. It is therefore necessary to do a temporal interest-lowering process on the text that has been historically clicked and browsed by the user, and first obtain the historical browsing time between when each text in the historical browsing text was clicked and browsed to the current time, specifically, using Representing text +.>Is at/>Click-through before day, i.e. historical browsing time is +.>Acquiring a first pearson correlation coefficient between the second document vector and the first document vector, performing interest weight reduction processing on the first pearson correlation coefficient according to the historical browsing time to acquire a second pearson correlation coefficient, and acquiring the preference degree of the target user according to the second pearson correlation coefficient, wherein the user preference degree of the first candidate text and the second candidate text is finally expressed as

In this embodiment, the historical browsing time from when each text in the historical browsing text is clicked to browse to the current time is obtained; acquiring a first pearson correlation coefficient between the second document vector and the first document vector, and performing interest weight reduction processing on the first pearson correlation coefficient according to the historical browsing time to acquire a second pearson correlation coefficient; and obtaining the preference degree of the target user according to the second pearson correlation coefficient. According to the embodiment, the preference degree of the target user is accurately obtained, and a foundation is laid for accurate recommendation.

Further, on the basis of the foregoing embodiment, in another embodiment of the present invention, the step of screening the selected text from the first candidate text and the second candidate text according to the propagation amount, the relevance, and the preference, and recommending the selected text to the target user includes:

Step D1, calculating the value score of each text in the first candidate text and the second candidate text according to the propagation quantity, the first correlation degree, the second correlation degree and the preference degree;

and D2, sequentially selecting a preset number of texts as selected texts according to the value score from high to low, and recommending the selected texts to the target user.

Selecting a selected text from the first candidate text and the second candidate text according to the propagation quantity, the first relevance, the second relevance and the user preference, and pushing the selected text to a target user, specifically, through a preset calculation formula such asAnd obtaining the value score of each text in the first candidate text and the second candidate text, sequentially selecting a preset number of texts as selected texts according to the value score from high to low, recommending the selected texts to the target user, and pushing 10 news with the largest score to the target user as target content, wherein the pushing can be carried out once per day.

In this embodiment, calculating a value score of each text in the first candidate text and the second candidate text according to the propagation amount, the first relevance, the second relevance and the preference; and sequentially selecting a preset number of texts as selected texts according to the value score from high to low, and recommending the selected texts to the target user. In this embodiment, accurate text recommendation is performed according to the value score.

Referring to fig. 3, fig. 3 is a schematic device structure diagram of a hardware running environment according to an embodiment of the present invention.

The text recommendation equipment in the embodiment of the invention can be a PC, or can be terminal equipment such as a smart phone, a tablet personal computer, a portable computer and the like.

As shown in fig. 3, the text recommendation device may include: a processor 1001, such as a CPU, memory 1005, and a communication bus 1002. Wherein a communication bus 1002 is used to enable connected communication between the processor 1001 and a memory 1005. The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also optionally be a storage device separate from the processor 1001 described above.

Optionally, the text recommendation device may further include a target user interface, a network interface, a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, and the like. The target user interface may comprise a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the selectable target user interface may further comprise a standard wired interface, a wireless interface. The network interface may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface).

It will be appreciated by those skilled in the art that the text recommendation device structure shown in FIG. 3 is not limiting of the text recommendation device and may include more or fewer components than shown, or may combine certain components, or a different arrangement of components.

As shown in fig. 3, an operating system, a network communication module, and a text recommendation program may be included in a memory 1005, which is one type of computer storage medium. An operating system is a program that manages and controls the hardware and software resources of the text recommendation device, supporting the execution of text recommendation programs and other software and/or programs. The network communication module is used to enable communication between components within the memory 1005 and other hardware and software in the text recommendation device.

In the text recommendation device shown in fig. 3, the processor 1001 is configured to execute a text recommendation program stored in the memory 1005, and implement the steps of the text recommendation method described in any one of the above.

The specific implementation manner of the text recommendation device of the present invention is basically the same as that of each embodiment of the text recommendation method, and will not be repeated here.

In addition, the embodiment of the invention also provides a text recommending device, which comprises:

Optionally, the text recommendation device further includes:

Optionally, the generating module includes:

Optionally, the selecting module includes:

Optionally, the screening module includes:

Optionally, the second acquisition unit includes:

Optionally, the seventh acquisition subunit is configured to implement:

Optionally, the eighth obtaining subunit is configured to implement:

Optionally, the screening module includes:

and the screening unit is used for sequentially selecting a preset number of texts from high to low according to the value score to serve as selected texts, and recommending the selected texts to the target user. The specific implementation manner of the text recommendation device is basically the same as that of each embodiment of the text recommendation method, and is not repeated here.

In addition, the embodiment of the invention also provides text recommendation equipment, which comprises: the computer-readable medium includes a memory 109, a processor 110, and a text recommendation program stored in the memory 109 and executable on the processor 110, which when executed by the processor 110, implements the steps of the various embodiments of the text recommendation method described above.

Furthermore, the present invention provides a computer medium storing one or more programs, where the one or more programs are further executable by one or more processors to implement the steps of the embodiments of the text recommendation method.

The expansion content of the specific implementation manner of the device and the medium (i.e. the computer medium) of the present invention is basically the same as that of each embodiment of the text recommendation method, and will not be described herein.

It should be noted that in this document, 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 one … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.

From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) comprising instructions for causing a terminal (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method according to the embodiments of the present invention.

The embodiments of the present invention have been described above with reference to the accompanying drawings, but the present invention is not limited to the above-described embodiments, which are merely illustrative and not restrictive, and many forms may be made by those having ordinary skill in the art without departing from the spirit of the present invention and the scope of the claims, which are to be protected by the present invention.

Claims

1. A text recommendation method, characterized in that the text recommendation method comprises:

generating a preset rational map based on the node text relation edges between the convergent node text and the convergent node text;

the preset event map is called, updated texts with the total association degree with the first candidate texts not smaller than a preset association threshold value are selected from the preset text database set according to the preset event map to serve as second candidate texts, and the event map comprises association relations between texts, wherein each association relation has the corresponding association degree;

2. The text recommendation method as claimed in claim 1, wherein the step of retrieving a preset rational map, selecting, as the second candidate text, an updated text having a total association degree with the first candidate text not smaller than a preset association threshold from the preset text database set according to the preset rational map comprises:

3. The text recommendation method as claimed in any one of claims 1-2, wherein said step of screening selected text from said first candidate text and said second candidate text and recommending said selected text to said target user according to said operation behavior comprises:

4. The text recommendation method of claim 3, wherein the step of obtaining a relevance of each text to the target user comprises:

5. The text recommendation method of claim 3, wherein the step of obtaining the preference degree of the target user according to the operation behavior comprises:

6. The text recommendation method of claim 5, wherein the step of obtaining a first document vector for each text in the historically browsed text comprises:

7. The text recommendation method of claim 5, wherein the step of obtaining a first pearson correlation coefficient between the second document vector and the first document vector, and obtaining the preference of the target user according to the first pearson correlation coefficient comprises:

8. The text recommendation method of claim 4, wherein the steps of screening selected text from the first candidate text and the second candidate text according to the propagation amount, the relevance, and the preference, and recommending the selected text to the target user include:

9. A text recommendation device, characterized in that the text recommendation device comprises:

the selecting module is used for acquiring texts to be processed from the preset text database set every interval preset time period; html tag filtering, symbol filtering and clause processing are carried out on the text to be processed through a preset regular expression, so that a preprocessed text formed by a clause list is obtained; identifying a plurality of preset text association relations for each clause in the clause list to obtain a node text to be processed, wherein the preset text association relations comprise, but are not limited to, compliance, causality, conditions and parallel relations; performing word segmentation on the node text to be processed through a preset word segmentation tool, acquiring word vectors of each word segment, and acquiring node vectors of each node text to be processed based on the word vectors of each word segment; calculating a first node distance between each node text to be processed and other node texts to be processed according to the node vector of each node text to be processed; performing iterative grafting treatment on two node texts to be treated, wherein the node distance of the node texts is smaller than a first preset distance, until each node text to be treated is in a convergence state that the node text relation edge is not changed any more, and setting each node text to be treated in the convergence state as a convergence node text; generating a preset rational map based on the node text relation edges between the convergent node text and the convergent node text; the preset event map is called, updated texts with the total association degree with the first candidate texts not smaller than a preset association threshold value are selected from the preset text database set according to the preset event map to serve as second candidate texts, and the event map comprises association relations between texts, wherein each association relation has the corresponding association degree;

10. A text recommendation device, the device comprising: memory, a processor and a text recommendation program stored on the memory and executable on the processor, which when executed by the processor, implements the steps of the text recommendation method according to any of claims 1 to 8.

11. A medium having stored thereon a text recommendation program which when executed by a processor implements the steps of the text recommendation method according to any of claims 1 to 8.