CN113254756A

CN113254756A - Advertisement recall method, device, equipment and storage medium

Info

Publication number: CN113254756A
Application number: CN202010088191.7A
Authority: CN
Inventors: 王雪琼; 徐奋飞; 徐伟; 曹海军; 冯峤
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2020-02-12
Filing date: 2020-02-12
Publication date: 2021-08-13
Anticipated expiration: 2040-02-12
Also published as: CN113254756B

Abstract

The embodiment of the application discloses an advertisement recall method, an advertisement recall device, advertisement recall equipment and a storage medium, and relates to the technical field of intelligent search. The specific implementation scheme is as follows: determining target entity nodes related to the historical search requests and the historical recalled advertisements in a target domain knowledge graph and target attribute nodes connected with the target entity nodes; determining the correlation between the historical search request and the historical recalled advertisement according to the target attribute node; and determining blacklist advertisements according to the correlation between the historical search requests and the historical recalled advertisements, and shielding the advertisements with the correlation lower than a threshold value when the advertisements are recalled. The target domain knowledge graph is applied to the advertisement recall of the search scene to detect the search request without the target domain correlation and the recalled advertisement in the historical search, so that the follow-up advertisement recall is effectively guided, irrelevant advertisements are completely shielded, the accuracy of the advertisement recall is improved, and the acquisition requirement of a user on the target domain information is met.

Description

Advertisement recall method, device, equipment and storage medium

Technical Field

The embodiment of the application relates to the technical field of computers, in particular to the technical field of intelligent search, and specifically relates to an advertisement recall method, an advertisement recall device, advertisement recall equipment and an advertisement recall storage medium.

Background

With the rapid development of the internet market, advertisements can be recalled for recommendation to users according to search requests initiated by users, wherein the advertisement recall in the medical field can be related, namely that medical related advertisements are recalled according to medical search intents in the search requests of the users. At present, all irrelevant advertisements cannot be effectively shielded by a conventional matching filtering mode, so that the accuracy of advertisement recall is low, and the user experience is influenced.

Disclosure of Invention

The embodiment of the application provides an advertisement recall method, an advertisement recall device, advertisement recall equipment and a storage medium, and the advertisement recall accuracy can be improved.

In a first aspect, an embodiment of the present application provides an advertisement recall method, including:

determining target entity nodes related to historical search requests and historical recalled advertisements in a target domain knowledge graph and target attribute nodes connected with the target entity nodes;

determining a correlation between the historical search request and the historical recalled advertisement according to the target attribute node;

and determining blacklist advertisements according to the correlation between the historical search request and the historical recalled advertisements, wherein the blacklist advertisements are used for shielding the advertisements with the correlation lower than a threshold value during advertisement recall.

One embodiment in the above application has the following advantages or benefits: the target domain knowledge graph is applied to the advertisement recall of the search scene to detect the search request without the target domain correlation and the recalled advertisement in the historical search, so that the follow-up advertisement recall is effectively guided, irrelevant advertisements are completely shielded, the accuracy of the advertisement recall is improved, and the acquisition requirement of a user on the target domain information is met.

Optionally, the determining target entity nodes associated with the historical search requests and the historical recalled advertisements in the target domain knowledge graph includes:

performing word segmentation on the historical search request and the target text in the historical recalled advertisement;

determining core words and candidate entity nodes of the target text according to the word segmentation result of the target text and the matching result between the nodes in the target domain knowledge graph;

and selecting a target entity node from the candidate entity nodes according to the core word.

One embodiment in the above application has the following advantages or benefits: the target entity nodes corresponding to the core words of the text are determined by segmenting the text of the historical search request and the historical recall advertisement and matching the segmented text with the target domain knowledge graph, so that the core entity content of the text can be obtained by mapping the target domain knowledge graph, and a basis is provided for determining the correlation between the historical search request and the historical recall advertisement.

Optionally, determining a core word of the target text according to a matching result between the word segmentation result of the target text and the node in the target domain knowledge graph, includes:

and selecting core words of the target text from the word segmentation results of the target text according to the word segmentation of the hit target field knowledge graph and the confidence coefficient in the target text.

One embodiment in the above application has the following advantages or benefits: based on the word segmentation result of the knowledge graph of the hit target field, the word segmentation method is favorable for selecting the word segmentation which can reflect the core content of the text most through the confidence coefficient of the word segmentation in the text environment.

Optionally, determining candidate entity nodes of the target text according to the matching result between the word segmentation result of the target text and the nodes in the target domain knowledge graph includes:

matching the core words with natural language word nodes in the target domain knowledge graph; wherein the natural language term nodes are trigger matching term representations of the nodes in the target domain knowledge graph;

and taking the entity node associated with the core word as a candidate entity node of the target text according to the matching result of the natural language word node and the connection relation between the entity node and the attribute node in the target domain knowledge graph.

One embodiment in the above application has the following advantages or benefits: the knowledge graph is matched and triggered through natural language word nodes in the knowledge graph in the target field, the natural language word nodes are used as the minimum matching granularity, the entity nodes of the root nodes are obtained through bottom-to-top matching based on the connection relation between the nodes, multiple representations of the entity nodes or attribute nodes in the natural language can be included in the maximum range, the matching strength of the knowledge graph is further improved, and the condition that the knowledge graph cannot be matched is avoided.

Optionally, after determining the blacklist advertisement according to the correlation between the historical search request and the historical recalled advertisement, the method further includes:

determining candidate recalled advertisements according to the received current search request;

performing word segmentation processing on the current search request, and reconstructing according to word segmentation results to obtain candidate search corpora;

and if the current search request or the candidate search corpus is detected to be the same as the historical search request, matching the candidate recalled advertisement with the blacklist advertisement so as to shield the candidate recalled advertisement hitting the blacklist advertisement.

One embodiment in the above application has the following advantages or benefits: through the determination of the blacklist advertisements, when the advertisements are recalled online, the current search request and the deformation form of the current search request are matched with the historical search request, so that the blacklist advertisements during periodic detection are shielded to the greatest extent, irrelevant or low-relevance advertisements are prevented from being recommended to a user, and the advertisement recall accuracy and the user experience are improved.

Optionally, the target domain knowledge graph is a medical domain knowledge graph.

One embodiment in the above application has the following advantages or benefits: in a medical scene, the application of the medical knowledge graph in advertisement search is beneficial to providing advertisements which are strongly correlated with medical contents searched by users for the users, and the error recall of the advertisements caused by common medical attributes is avoided.

Optionally, before the determining the target entity node associated with the historical search request and the historical recalled advertisement in the target domain knowledge graph and the target attribute node connected to the target entity node, further includes:

and according to the acquired disease information, constructing a medical knowledge graph with the entity nodes as the core by taking the disease as the entity nodes and taking other disease information as the attribute nodes.

One embodiment in the above application has the following advantages or benefits: by constructing the medical knowledge graph with the disease as the entity node and other disease information as the attribute node, any disease information is matched to obtain a corresponding disease entity, so that the detection of correlation is performed based on the disease entity.

Optionally, the constructing of the medical knowledge graph with the entity node as the core includes:

mining the nodes in the medical knowledge graph according to the synonymy relationship;

and fusing different entity nodes according to at least one of the name attribute, the alias attribute and the synonymy relationship of the entity nodes to obtain the redundancy-removed medical knowledge graph.

One embodiment in the above application has the following advantages or benefits: in the process of constructing the medical knowledge graph, the nodes are mined according to the synonymy relationship, so that the method is beneficial to determining various word expressions of the same node. Meanwhile, by fusing different entity nodes in the medical knowledge graph and reserving the fused entity nodes in other attribute forms, the matching range of the knowledge graph is enlarged, and the redundancy of the medical knowledge graph is removed.

Optionally, the determining, according to the target attribute node, a correlation between the historical search request and the historical recalled advertisement includes:

determining the correlation between the historical search request and the historical recalled advertisement according to a core attribute node connected with the target entity node; wherein the core attribute nodes comprise at least a name attribute, a department attribute, and a symptom attribute.

One embodiment in the above application has the following advantages or benefits: in a medical scenario, core attribute nodes are taken as nodes capable of absolutely distinguishing disease entities, and correlation between historical search requests and historical recall advertisements is favorably determined based on the core attribute nodes.

In a second aspect, an embodiment of the present application provides an advertisement recall apparatus, including:

the node matching module is used for determining a target entity node associated with the historical search request and the historical recall advertisement in a target domain knowledge graph and a target attribute node connected with the target entity node;

an advertisement relevance determination module, configured to determine, according to the target attribute node, a relevance between the historical search request and the historical recalled advertisement;

and the advertisement shielding module is used for determining the blacklist advertisement according to the correlation between the historical search request and the historical recalled advertisement and shielding the advertisement with the correlation lower than a threshold value during advertisement recall.

In a third aspect, an embodiment of the present application provides an electronic device, including:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform an advertisement recall method as described in any embodiment of the present application.

In a fourth aspect, embodiments of the present application provide a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform an advertisement recall method according to any of the embodiments of the present application.

One embodiment in the above application has the following advantages or benefits: and determining entity nodes related to the historical search request and the historical recalled advertisements based on the target domain knowledge graph according to the historical search request and the historical recalled advertisements, and determining the correlation between the historical search request and the historical recalled advertisements according to the related entity nodes, so that the historical recalled advertisements with the correlation lower than a threshold value are added into the blacklist, and the advertisements in the blacklist are shielded when the same search request is initiated again. According to the method and the device, the target field knowledge graph is applied to the advertisement recall of the search scene, so that the search request without the target field correlation in the historical search and the recalled advertisement are detected, the follow-up advertisement recall is effectively guided, irrelevant advertisements are completely shielded, the advertisement recall accuracy is improved, and the requirement of a user for acquiring the target field information is met.

Other effects of the above-described alternative will be described below with reference to specific embodiments.

Drawings

The drawings are included to provide a better understanding of the present solution and are not intended to limit the present application. Wherein:

FIG. 1 is a flow chart of a method of advertisement recall according to a first embodiment of the present application;

FIG. 2 is a flow chart of a method of advertisement recall according to a second embodiment of the present application;

FIG. 3 is a partial illustration of a medical knowledge-map according to a second embodiment of the present application;

FIG. 4 is a block diagram of a medical knowledge-graph based advertising recall in accordance with a second embodiment of the present application;

FIG. 5 is a flowchart of an advertisement recall method according to a third embodiment of the present application;

FIG. 6 is a flowchart of an advertisement recall method according to a fourth embodiment of the present application;

fig. 7 is a schematic structural diagram of an advertisement recall apparatus according to a fifth embodiment of the present application;

FIG. 8 is a block diagram of an electronic device for implementing an advertisement recall method of an embodiment of the present application.

Detailed Description

The following description of the exemplary embodiments of the present application, taken in conjunction with the accompanying drawings, includes various details of the embodiments of the application for the understanding of the same, which are to be considered exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present application. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

First embodiment

Fig. 1 is a flowchart of an advertisement recall method according to a first embodiment of the present application, which is applicable to periodically detecting correlations between a historical search request initiated by a user and a recalled historical recalled advertisement thereof based on a target domain knowledge map, and determining a blacklisted advertisement having a correlation below a threshold value, so as to mask out irrelevant advertisements in online advertisement recall. As shown in fig. 1, the method specifically includes the following steps:

s110, determining target entity nodes related to the historical search requests and the historical recalled advertisements in the target domain knowledge graph and target attribute nodes connected with the target entity nodes.

In particular embodiments of the present application, the target domain may be any domain related to a search, such as a medical domain or the like. The historical search request refers to a search request initiated in a target domain by a large number of users in a network in the past period of time. Accordingly, the historical recalled advertisements refer to advertisements that the advertisement platform has recalled in response to historical search requests. Wherein, the different historical recall advertisements have different correlations with the same historical search request, and the stronger the correlation is, the more the historical recall advertisement meets the historical search requirement of the user.

In the embodiment, aiming at the technical problem that all irrelevant advertisements cannot be effectively shielded when the advertisements are recalled only based on the keyword precise matching in the prior art, the target domain knowledge graph is applied to the advertisement recall according to the search request by constructing the target domain knowledge graph. The target domain knowledge graph at least comprises entity nodes and attribute nodes, and the entity nodes are used as cores, and the attribute nodes are used as concrete explanation descriptions of the entity nodes, so that the associated entity nodes can be matched according to the attribute nodes. Wherein different entity nodes may have the same attribute node.

In addition, the target domain knowledge graph can further comprise natural language word nodes, the natural language word nodes are in one-to-one correspondence with the entity nodes or attribute nodes in the target domain knowledge graph and are word representations of different forms of the corresponding entity nodes or attribute nodes, multiple representations of the entity nodes or attribute nodes in the natural language can be contained in the largest range, and the natural language word nodes or attribute nodes can be used as the minimum matching granularity of the target domain knowledge graph and used for triggering the matching of the nodes so as to improve the matching strength of the knowledge graph and avoid the condition that the nodes cannot be matched.

Illustratively, in the medical field, the nodes in the medical knowledge graph are determined by taking diseases as entity nodes and other disease information as attribute nodes according to disease information collected from the website of the medical institution. Respectively carrying out synonymy relation mining on entity nodes and attribute nodes in the medical knowledge graph; and fusing different entity nodes according to at least one of the name attribute, the alias attribute and the synonymy relationship of the entity nodes to obtain the redundancy-free medical knowledge graph containing the dimension information. The natural language word nodes as the corresponding nodes can be represented by folk words of various diseases or attributes. Therefore, matching can be triggered based on the natural language word nodes, attribute nodes are determined according to the matched natural language word nodes, and entity nodes connected with the attribute nodes are determined according to the connection relation among the nodes in the target domain knowledge graph.

Specifically, at least one of a title, a text or a picture character in the historical search request and the historical recall advertisement can be used as a target text to be matched, word segmentation processing is performed on the target text, each obtained word is matched with a node in the target domain knowledge graph, and a core word and a candidate entity node which hit the target domain knowledge graph and can embody the core content of the target text in the target text are determined. The core words of the target text can be selected from the word segmentation results of the target text according to the word segmentation of the hit target field knowledge graph and the confidence coefficient in the target text. Matching the core words with natural language word nodes in the target domain knowledge graph, and taking entity nodes associated with the core words as candidate entity nodes of the target text. Finally, based on the correlation model, the correlation degree between the core word and the candidate entity nodes is calculated, the candidate entity nodes with lower correlation degree are filtered, and the candidate entity nodes with the highest correlation degree and meeting a certain threshold value are used as final target entity nodes.

And S120, determining the correlation between the historical search request and the historical recalled advertisement according to the target attribute node.

In the embodiment of the present application, after determining the target entity node of the historical search request and the target attribute node associated therewith, and the target entity node of the historical recalled advertisement and the target attribute node associated therewith, the correlation between the historical search request and the historical recalled advertisement may be determined based on the target attribute nodes of the historical search request and the historical recalled advertisement.

For example, the attribute nodes of the entity node may be divided into core attribute nodes and non-core attribute nodes in advance based on the characteristics of the entity in the target domain. The core attribute node is generally the mark attribute of the entity node, and has better distinguishability on the entity; non-core attribute nodes are some attributes that are not so distinct. Therefore, the correlation between the historical search request and the historical recalled advertisement is determined according to the core attribute node connected with the target entity node.

Illustratively, in the medical field, the atlas structure of the example of cold illness is shown in table 1. The core attribute nodes at least comprise name attributes, department attributes and symptom attributes. Suppose that the entity node targeted by the historical search request is determined to be A and the entity node targeted by a certain historical recalled advertisement is B. Based on the medical knowledge graph, if department attribute nodes of a and B are detected to be the same or have a relationship of upper and lower levels, or if a and B are detected to be the same disease entity node or the department attribute nodes of a and B have an intersection on department attributes, or if a department of a disease is detected to be contained in department information of another disease, it may be determined that the history search request has a correlation with the history recall advertisement.

TABLE 1 example chart structure

Item	Value of	Sample examples
			Class one (category)	First-class department	Internal medicine
Second order (category)	Secondary department	Respiratory medicine
			Entity (disease)	Signature of disease name
Core Attribute (core _ attr)	Name of disease	Common cold
			Core Attribute (core _ attr)	Symptoms and signs	Sneezing
Core Attribute (core _ attr)	Department's office	Respiratory medicine
			Non-core Properties (nocore _ attr)	Location of a body part	Nose
Non-core Properties (nocore _ attr)	Administration of drugs	Radix Isatidis
			Non-core Properties (nocore _ attr)	Clinical examination	Stenosis of nasal cavity
Non-core Properties (nocore _ attr)	Susceptible population	Children's toy
			Non-core Properties (nocore _ attr)	Mode of infection	Droplet propagation
Non-core Properties (nocore _ attr)	Method of treatment	Medicine for treating diseaseTherapy
			Natural language word node (mention)	Value and synonyms of attributes

S130, according to the correlation between the historical search request and the historical recalled advertisements, determining the blacklist advertisements, and shielding the advertisements with the correlation lower than a threshold value when the advertisements are recalled.

In an embodiment of the present application, a blacklist advertisement refers to a history recall advertisement having no correlation with a history search request or having a correlation lower than a threshold, and is used to instruct a retrieval system to initiate a search and an advertisement recall again according to a recombination of the history search request or a segmentation result of the history search request, and if a recall result includes an advertisement in the blacklist, the advertisement is blocked online to avoid recommending an advertisement with a lower correlation to a user.

Specifically, by periodically performing entity linking in S110 and relevance discrimination in S120, a historical recall advertisement having a relevance below a threshold value determined based on target domain knowledge graph detection is added to the blacklist. In the blacklist, the incidence relation between the historical search request and the recombination of the segmentation result of the historical search request and the blacklist advertisement can be established.

According to the technical scheme, the historical search request and the recalled historical recalled advertisements are used as the basis, the entity nodes related to the historical search request and the historical recalled advertisements are determined based on the target domain knowledge graph, the correlation between the historical search request and the historical recalled advertisements is determined according to the related entity nodes, and therefore the historical recalled advertisements with the correlation lower than the threshold value are added to the blacklist, and the advertisements in the blacklist are shielded when the same search request is initiated again. According to the method and the device, the target field knowledge graph is applied to the advertisement recall of the search scene, so that the search request without the target field correlation in the historical search and the recalled advertisement are detected, the follow-up advertisement recall is effectively guided, irrelevant advertisements are completely shielded, the advertisement recall accuracy is improved, and the requirement of a user for acquiring the target field information is met.

Second embodiment

Fig. 2 is a flowchart of an advertisement recall method according to a second embodiment of the present application, and this embodiment further explains, on the basis of the first embodiment, the determination of the historical search request and the entity node associated with the historical recall advertisement based on the target domain knowledge graph, so that the entity node can be obtained by matching based on the text core word determined by the word segmentation result of the text. As shown in fig. 2, the method specifically includes the following steps:

and S210, performing word segmentation on the historical search request and the target text in the historical recall advertisement.

In the specific embodiment of the present application, the historical search request may be used as a target text, and the word segmentation processing may be performed on the historical search request text. The word segmentation processing can be carried out on the historical recalled advertisement text by taking the advertisement title of the historical recalled advertisement, or the text content in the historical recalled advertisement, even the text content identified from the advertisement picture as the target text. In this embodiment, the word segmentation processing method is not limited, and any method capable of performing word segmentation processing on a text may be applied to this embodiment.

For example, assuming that the historical search request is "what medicine is eaten by cold," the text of the historical search request may be used as the target text, and the word segmentation processing is performed on the target text to obtain word segmentation results "cold", "cough", "eating", "what", and "medicine".

S220, determining core words and candidate entity nodes of the target text according to the word segmentation result of the target text and the matching result between the nodes in the target field knowledge graph.

In the specific embodiment of the application, after the word segmentation processing is performed on the target text, each word in the word segmentation result of the target text is matched with a node in the target domain knowledge graph, and on the basis of the matching result between the word segmentation result and the node in the target domain knowledge graph, subsequent entity linking and relevance judgment are performed. In order to improve the matching capability of the target domain knowledge graph, the natural language word nodes are used as the minimum matching granularity and used for triggering the matching of the nodes.

Illustratively, taking a medical knowledge-graph as an example, fig. 3 is a partial example of the medical knowledge-graph. As shown in fig. 3, disease is used as an entity node, name, cough, symptom, medication, clinical examination, treatment method, susceptible population, and propagation method are used as attribute nodes, and a comment tag is used as a natural language word node. For example, when matching is performed based on the names of diseases, no matter whether the participle in the target text is represented by multiple words such as cold, common cold or catarrhal rhinitis, the corresponding name attribute node cold and the disease entity node connected with the attribute node can be matched based on the natural language word node.

In this embodiment, the core word refers to a word segment that can best reflect the core content of the target text. Optionally, the core word of the target text is selected from the word segmentation result of the target text according to the word segmentation of the hit target field knowledge graph and the confidence in the target text. In this embodiment, the determination method of the confidence level is not limited, and any method capable of determining the confidence level may be applied to this embodiment.

Illustratively, according to a matching result between a word segmentation result of a target text and a node in a target domain knowledge graph, for a word of a hit target domain knowledge graph, calculating the confidence of the word of the hit target domain knowledge graph in the target text based on wordrank technology, sequencing the words according to the confidence, and taking the word with the highest confidence as a core word of the target text. For example, in the above example, in the word segmentation result, the words "cold", "cough", and "medicine" hit the medical knowledge map. And calculating the confidence degrees of the segmented words of 'cold', 'cough' and 'medicine' in the target text based on the wordrank technology to obtain the highest confidence degree of the segmented word of 'cold' in the context of the historical search request, so that the segmented word of 'cold' is used as the core word of the historical search request.

In this embodiment, the candidate entity node is an entity node that can relatively represent the core content of the target text, among all entity nodes associated with the participles of the target domain knowledge graph hit by the target text, and the entity node matched with the core word in the target domain knowledge graph can be used as the candidate entity node of the target text. Optionally, matching the core words with natural language word nodes in the target domain knowledge graph; and taking the entity nodes associated with the core words as candidate entity nodes of the target text according to the matching result of the natural language word nodes and the connection relation between the entity nodes and the attribute nodes in the target domain knowledge graph. And according to the matching of the target domain knowledge graph, more than one entity node associated with the core word can be provided. For example, individual symptoms of different diseases may be the same, and thus when the symptoms are used as core words for matching, different disease entity nodes can be obtained through matching.

In this embodiment, in order to improve the entity link efficiency and avoid the invalid processing of the historical recalled advertisements that do not relate to the target domain knowledge graph, the historical recalled advertisements to which the target texts of the target domain knowledge graph belong may be filtered out when matching is performed based on natural language word nodes after the word segmentation results of the target texts of the historical recalled advertisements are determined.

And S230, selecting a target entity node from the candidate entity nodes according to the core words.

In the specific embodiment of the present application, the target entity node refers to an entity node that can best embody the core content of the target text. The relevance between the core words and the candidate entity nodes can be calculated based on the relevance model, the candidate entity nodes with low relevance are filtered out, and the candidate entity nodes with highest relevance and meeting a certain threshold value are used as final target entity nodes.

Illustratively, in the above example, in the word segmentation of the hit medical knowledge graph, it is assumed that the entity node associated with the word segmentation of "cold" includes A, B and C, the entity node associated with the word segmentation of "cough" includes D and E, and the entity node associated with the word segmentation of "medicine" includes F. And determining the participle 'cold' as a core word according to the confidence of each hit participle in the target text, and taking the entity nodes A, B and C as candidate entity nodes. And calculating the correlation degree between the core word and the candidate entity node based on the correlation model, and taking the candidate entity node which has the highest correlation degree in A, B and C and meets a certain threshold value as a final target entity node.

S240, according to the target attribute node connected with the target entity node, determining the correlation between the historical search request and the historical recall advertisement.

And S250, determining the blacklist advertisement according to the correlation between the historical search request and the historical recalled advertisement, and shielding the advertisement with the correlation lower than a threshold value when the advertisement is recalled.

For example, taking the medical field as an example, fig. 4 is a block diagram of an advertisement recall based on a medical knowledge graph. As shown in fig. 4, disease information of an authoritative medical website, a phoenix nest medical material, and a DMP (Data Management Platform) is used as a disease Data source. In the medical knowledge map platform, a medical knowledge map is constructed through entity construction, synonymy relation mining, entity fusion and natural language word node construction based on a disease data source to form a disease knowledge Base (KG-Base). Thus, when the historical recall advertisements are periodically detected, the medical knowledge map platform performs entity linking on the target text through text matching filtering, entity recognition, wordrank and relevance filtering. And finally, matching is carried out based on disease entity nodes linked by the historical search request and the historical recalled advertisements, and the correlation between the historical search request and the historical recalled advertisements is determined so as to mine the blacklist advertisements, wherein the blacklist advertisements are used for shielding the recalled blacklist advertisements during advertisement recall.

According to the technical scheme, word segmentation is carried out on the historical search request and the target text of the recalled historical recalled advertisement, node matching is carried out based on a target domain knowledge graph, the historical search request, the core words and the candidate entity nodes of the historical recalled advertisement are determined, the target entity nodes are selected from the candidate entity nodes according to the core words, the correlation between the historical search request and the historical recalled advertisement is determined according to the target attribute nodes connected with the target entity nodes, and therefore the historical recalled advertisement with the correlation lower than the threshold value is added to the blacklist, and the advertisement in the blacklist is shielded when the same search request is initiated again. According to the method and the device, the target field knowledge graph is applied to the advertisement recall of the search scene, so that the search request without the target field correlation in the historical search and the recalled advertisement are detected, the follow-up advertisement recall is effectively guided, irrelevant advertisements are completely shielded, the advertisement recall accuracy is improved, and the requirement of a user for acquiring the target field information is met.

Third embodiment

Fig. 5 is a flowchart of an advertisement recall method according to a third embodiment of the present application, and this embodiment further explains the correlation detection between the historical search request and the historical recall advertisement in the medical field on the basis of the first embodiment, so as to construct a medical knowledge graph based on disease information and perform correlation detection based on the medical knowledge graph. As shown in fig. 5, the method specifically includes the following steps:

s510, according to the collected disease information, the disease is used as an entity node, other disease information is used as an attribute node, and a medical knowledge graph with the entity node as a core is constructed.

In the specific embodiment of the present application, in the medical field, the disease information is information related to diseases, such as cases, papers, etc., collected or crawled in advance from websites of authoritative medical institutions. Through the modes of keyword identification and the like, the words such as diseases, attributes and the like in the disease information can be identified, and the nodes in the medical knowledge graph can be determined. And according to professional knowledge in the medical field, edges among the nodes are established, and the association relation among the nodes is determined.

Optionally, mining the synonymy relation of the nodes in the medical knowledge graph; and fusing different entity nodes according to at least one of the name attribute, the alias attribute and the synonymy relationship of the entity nodes to obtain the redundancy-removed medical knowledge graph.

In this embodiment, the indication of disease information is relatively professional, and professional indications of the same word may include more than one type, given that the source of the disease information is relatively authoritative. Therefore, it is necessary to mine nodes having synonymous relationships in the medical knowledge graph. For example, for the attribute node, the synonymy relationship mining of the attribute name may be performed on the attribute node in the graph according to a synonym table of the predetermined synonymy relationship attribute word, so as to determine the attribute node having the synonymy relationship, and expand various professional expression forms of the attribute node. For entity nodes, synonymy relationship mining can be performed on entities in the graph according to the text similarity of disease names and/or the number of common attributes associated with the two entities. For example, rhinitis and chronic rhinitis, although professionally different, are almost no greater for the user.

In this embodiment, the mining of the synonymous relationship realizes the sufficient extension of the professional knowledge of the medical knowledge graph, and in order to improve the simplification and the application efficiency of the medical knowledge graph, the entity nodes need to be fused. Specifically, different entities and nodes satisfying the following arbitrary relationships are fused: the entity names are the same; the alias attributes of the entity nodes have the same alias; two entities having a synonymous relationship. When the entity nodes are fused, the entity nodes with more attribute labels are used as the fused entity nodes, the union set of the associated attributes of the fused entity nodes is reserved, and the fused entity nodes are reserved in the form of alias attributes.

In addition, after the medical knowledge map is constructed based on medical professional knowledge, the natural language representation form of each node can be extracted according to folk representation and used as the natural language word node of each node. And furthermore, the professional requirement of the user for inputting a medical search request is reduced, the user can initiate a search in any word representation known by the user, and the medical knowledge graph can be matched according to any word representation.

S520, performing word segmentation processing on the historical search request and the target text in the historical recall advertisement.

S530, selecting core words of the target text from the word segmentation results of the target text according to the degree of confidence of the hit medical knowledge map in the target text.

S540, matching the core words with natural language word nodes in the medical knowledge graph; wherein the natural language term nodes are the trigger matching term representations of the nodes in the medical knowledge graph.

And S550, taking the entity node associated with the core word as a candidate entity node of the target text according to the matching result of the natural language word node and the connection relation between the entity node and the attribute node in the medical knowledge graph.

And S560, selecting a target entity node from the candidate entity nodes according to the core words.

S570, determining the correlation between the historical search request and the historical recall advertisement according to the core attribute node connected with the target entity node; the core attribute nodes at least comprise name attributes, department attributes and symptom attributes.

S580, according to the correlation between the historical search request and the historical recalled advertisements, determining the blacklist advertisements, and shielding the advertisements with the correlation lower than a threshold value when the advertisements are recalled.

According to the technical scheme, the medical knowledge graph which takes the diseases as the entity nodes and takes other disease information as the attribute nodes is constructed and applied to the advertisement recall of the search scene, so that the search request without medical relevance and the recalled advertisements in the historical search logs are detected, the follow-up advertisement recall is effectively guided, irrelevant advertisements are completely shielded, the accuracy of the advertisement recall is improved, and the acquisition requirements of users on the medical information are met.

Fourth embodiment

Fig. 6 is a flowchart of an advertisement recall method according to a fourth embodiment of the present application, and this embodiment further explains, based on the first embodiment, an online advertisement recall process based on blacklist advertisements, which can mask out advertisements with relevance lower than a threshold value based on blacklist advertisements. As shown in fig. 6, the method specifically includes the following steps:

s610, determining target entity nodes related to the historical search requests and the historical recall advertisements in the target domain knowledge graph and target attribute nodes connected with the target entity nodes.

S620, according to the target attribute node, determining the correlation between the historical search request and the historical recall advertisement.

S630, according to the correlation between the historical search request and the historical recall advertisement, determining the blacklist advertisement, and shielding the advertisement with the correlation lower than a threshold value when the advertisement is recalled.

In an embodiment of the present application, relevance determination may be performed on historical search requests and historical recalled advertisements recalled thereof periodically to determine blacklisted advertisements. Illustratively, entity linking and relevance determination is performed at night-time each day, targeting the entire historical search requests of all users of the day and the advertisement title of the recalled historical recalled advertisement. Historical search requests with no or reduced relevance and their associated historical recalled advertisements are added to the blacklist. For example, historical search requests that are subject to colds and historical recall advertisements that are recalled from them that are subject to reshaping, and the associations between the two, are added to the blacklist. And when any user initiates a search for the cold, if the recalled advertisements comprise advertisement integer types, namely hit blacklists, the recalled integer advertisements are filtered.

And S640, determining candidate recalled advertisements according to the received current search request.

In the specific embodiment of the present application, the current search request refers to a search request initiated by a user and received online in real time. Accordingly, based on any advertisement recall manner in the prior art, all advertisements recalled according to the current search request are taken as candidate recalled advertisements to be recommended to the user.

S650, performing word segmentation processing on the current search request, and reconstructing according to word segmentation results to obtain candidate search corpora.

In the specific embodiment of the application, because the text with the same semantic meaning has various expression modes and different expression habits of different users, the word segmentation processing is performed on the current search request, and the reconstruction is performed based on the word segmentation result to obtain various expression forms for expressing the current search request, so that candidate search corpora are formed.

S660, if the current search request or the candidate search corpus is detected to be the same as the historical search request, matching the candidate recall advertisement with the blacklist advertisement so as to shield the candidate recall advertisement hit on the blacklist advertisement.

In the specific embodiment of the application, the current search request and the candidate search corpus are matched with the historical search request, and whether the same or similar search requests are initiated historically is detected. And if the current search request or the candidate search corpus is detected to be the same as the historical search request, matching the candidate recall advertisements with the blacklist advertisements, and shielding the candidate recall advertisements hitting the blacklist advertisements under the guidance of the blacklist advertisements. And further, the defect that all irrelevant advertisements cannot be effectively shielded in the prior art is overcome.

According to the technical scheme, the current search request and the deformation form of the current search request are matched with the historical search request when the advertisements are recalled online through the determination of the blacklist advertisements, the blacklist advertisements during periodic detection are shielded to the greatest extent, irrelevant advertisements or advertisements with low relevance are prevented from being recommended to a user, and the advertisement recall accuracy and the user experience are improved.

Fifth embodiment

Fig. 7 is a schematic structural diagram of an advertisement recall apparatus according to a fifth embodiment of the present application, which is applicable to periodically detecting a correlation between a historical search request initiated by a user and a recalled historical recall advertisement thereof based on a target domain knowledge graph, and determining a blacklist advertisement with the correlation lower than a threshold value, so as to mask off irrelevant advertisements during online advertisement recall. The apparatus 700 specifically includes the following:

a node matching module 710, configured to determine target entity nodes associated with the historical search requests and the historical recalled advertisements in the target domain knowledge graph, and target attribute nodes connected to the target entity nodes;

an advertisement relevance determination module 720, configured to determine, according to the target attribute node, a relevance between the historical search request and the historical recalled advertisement;

the advertisement shielding module 730 is configured to determine a blacklist advertisement according to a correlation between the historical search request and the historical recalled advertisement, and shield an advertisement with a correlation lower than a threshold value when the advertisement is recalled.

Optionally, the node matching module 710 is specifically configured to:

Further, the apparatus 700 further includes an advertisement online recall module 740, specifically configured to:

after determining the blacklist advertisement according to the correlation between the historical search request and the historical recalled advertisement, determining a candidate recalled advertisement according to the received current search request;

Further, the apparatus 700 further includes a map building module 750, specifically configured to:

before determining a target entity node associated with the historical search request and the historical recall advertisement in the target domain knowledge graph and a target attribute node connected with the target entity node, constructing a medical knowledge graph taking the entity node as a core according to the collected disease information and taking the disease as the entity node and other disease information as the attribute node.

Optionally, the map building module 750 is specifically configured to:

Optionally, the advertisement relevance determining module 720 is specifically configured to:

According to the technical scheme of the embodiment, through the mutual cooperation of all the functional modules, the functions of constructing a knowledge map of the target field, linking entities, judging the correlation between historical search requests and historical recalled advertisements, determining the blacklist, recalling the advertisements in real time on line, shielding the blacklist advertisements and the like are realized. According to the method and the device, the target field knowledge graph is applied to the advertisement recall of the search scene, so that the search request without the target field correlation in the historical search and the recalled advertisement are detected, the follow-up advertisement recall is effectively guided, irrelevant advertisements are completely shielded, the advertisement recall accuracy is improved, and the requirement of a user for acquiring the target field information is met.

Sixth embodiment

According to an embodiment of the present application, an electronic device and a readable storage medium are also provided.

Fig. 8 is a block diagram of an electronic device according to an advertisement recall method of an embodiment of the present application. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the present application that are described and/or claimed herein.

As shown in fig. 8, the electronic apparatus includes: one or more processors 801, memory 802, and interfaces for connecting the various components, including a high speed interface and a low speed interface. The various components are interconnected using different buses and may be mounted on a common motherboard or in other manners as desired. The processor may process instructions for execution within the electronic device, including instructions stored in or on the memory to display Graphical information for a Graphical User Interface (GUI) on an external input/output device, such as a display device coupled to the Interface. In other embodiments, multiple processors and/or multiple buses may be used, along with multiple memories and multiple memories, as desired. Also, multiple electronic devices may be connected, with each device providing portions of the necessary operations, e.g., as a server array, a group of blade servers, or a multi-processor system. Fig. 8 illustrates an example of a processor 801.

The memory 802 is a non-transitory computer readable storage medium as provided herein. Wherein the memory stores instructions executable by at least one processor to cause the at least one processor to perform the advertisement recall method provided herein. A non-transitory computer readable storage medium of the present application stores computer instructions for causing a computer to perform an advertisement recall method provided by the present application.

The memory 802, which is a non-transitory computer readable storage medium, may be used to store non-transitory software programs, non-transitory computer executable programs, and modules, such as program instructions/modules corresponding to the advertisement recall method in the embodiments of the present application, for example, the node matching module 710, the advertisement relevance determination module 720, the advertisement blocking module 730, the online advertisement recall module 740, and the graph construction module 750 shown in fig. 7. The processor 801 executes various functional applications of the server and data processing by running non-transitory software programs, instructions, and modules stored in the memory 802, that is, implements the advertisement recall method in the above-described method embodiments.

The memory 802 may include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function; the storage data area may store data created according to use of the electronic device of the advertisement recall method, and the like. Further, the memory 802 may include high speed random access memory and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, the memory 802 optionally includes memory located remotely from the processor 801, which may be connected to the electronic device of the advertisement recall method via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

The electronic device of the advertisement recall method may further include: an input device 803 and an output device 804. The processor 801, the memory 802, the input device 803, and the output device 804 may be connected by a bus or other means, and are exemplified by a bus in fig. 8.

The input device 803 may receive input numeric or character information and generate key signal inputs related to user settings and function control of the electronic apparatus for the advertisement recall method, such as an input device like a touch screen, a keypad, a mouse, a track pad, a touch pad, a pointer stick, one or more mouse buttons, a track ball, a joystick, etc. The output device 804 may include a display apparatus, an auxiliary lighting device such as a Light Emitting Diode (LED), a tactile feedback device, and the like; the tactile feedback device is, for example, a vibration motor or the like. The Display device may include, but is not limited to, a Liquid Crystal Display (LCD), an LED Display, and a plasma Display. In some implementations, the display device can be a touch screen.

Various implementations of the systems and techniques described here can be realized in digital electronic circuitry, Integrated circuitry, Application Specific Integrated Circuits (ASICs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.

These computer programs, also known as programs, software applications, or code, include machine instructions for a programmable processor, and may be implemented using high-level procedural and/or object-oriented programming languages, and/or assembly/machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and/or Device for providing machine instructions and/or data to a Programmable processor, such as a magnetic disk, optical disk, memory, Programmable Logic Device (PLD), including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal used to provide machine instructions and/or data to a programmable processor.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device for displaying information to a user, for example, a Cathode Ray Tube (CRT) or an LCD monitor; and a keyboard and a pointing device, such as a mouse or a trackball, by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback, e.g., visual feedback, auditory feedback, or tactile feedback; and input from the user may be received in any form, including acoustic, speech, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a back-end component, e.g., as a data server, or that includes a middleware component, e.g., an application server, or that includes a front-end component, e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here, or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication, e.g., a communication network. Examples of communication networks include: local Area Networks (LANs), Wide Area Networks (WANs), the internet, and blockchain networks.

The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.

According to the technical scheme of the embodiment of the application, the target domain knowledge graph is applied to the advertisement recall of the search scene to detect the search request without the target domain correlation and the recalled advertisement in the historical search, so that the follow-up advertisement recall is effectively guided, irrelevant advertisements are completely shielded, the advertisement recall accuracy is improved, and the acquisition requirement of a user on the target domain information is met.

In addition, the target entity nodes corresponding to the core words of the text are determined by segmenting the text of the historical search request and the historical recall advertisement and matching the segmented text with the target domain knowledge graph, so that the core entity content of the text can be obtained by mapping the target domain knowledge graph, and a basis is provided for determining the correlation between the historical search request and the historical recall advertisement.

In addition, the segmentation result of the target domain knowledge graph is hit as the basis, and the confidence of the segmentation in the text environment is favorable for selecting the segmentation which can reflect the core content of the text most.

In addition, matching triggering of the knowledge graph is carried out through natural language word nodes in the knowledge graph of the target field, the natural language word nodes are used as the minimum matching granularity, entity nodes of the root nodes are obtained through matching from bottom to top based on the connection relation between the nodes, multiple representations of the entity nodes or attribute nodes in the natural language can be contained in the maximum range, the matching strength of the knowledge graph is further improved, and the condition that matching cannot be carried out is avoided.

In addition, through the determination of the blacklist advertisements, when the advertisements are recalled online, the current search request and the deformation form of the current search request are matched with the historical search request, so that the blacklist advertisements during periodic detection are shielded to the greatest extent, irrelevant or low-relevance advertisements are prevented from being recommended to the user, and the advertisement recall accuracy and the user experience are improved.

In addition, in a medical scene, the application of the medical knowledge graph in advertisement searching is beneficial to providing advertisements which are strongly correlated with medical contents searched by users for the users, and the error recalling of the advertisements caused by common medical attributes is avoided.

In addition, by constructing the medical knowledge graph with the disease as the entity node and other disease information as the attribute node, any disease information is matched to obtain a corresponding disease entity, so that the detection of correlation is performed based on the disease entity.

In addition, in the process of constructing the medical knowledge graph, the nodes are mined according to the synonymy relationship, and the method is favorable for determining various word expressions of the same node. Meanwhile, by fusing different entity nodes in the medical knowledge graph and reserving the fused entity nodes in other attribute forms, the matching range of the knowledge graph is enlarged, and the redundancy of the medical knowledge graph is removed.

In addition, in a medical scenario, the core attribute node is taken as a node capable of absolutely distinguishing disease entities, and thus the relevance between the historical search request and the historical recall advertisement is favorably determined based on the core attribute node.

It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present application may be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solutions disclosed in the present application can be achieved, and the present invention is not limited herein.

The above-described embodiments should not be construed as limiting the scope of the present application. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An advertisement recall method, comprising:

2. The method of claim 1, wherein determining a target entity node for which historical search requests and historical recall advertisements are associated in a target domain knowledge graph comprises:

3. The method of claim 2, wherein determining the core words of the target text according to the matching result between the word segmentation result of the target text and the nodes in the target domain knowledge graph comprises:

4. The method of claim 2, wherein determining candidate entity nodes of the target text according to the matching result between the word segmentation result of the target text and the nodes in the target domain knowledge graph comprises:

5. The method of claim 1, further comprising, after said determining blacklisted advertisements based on relevance between said historical search requests and said historical recalled advertisements:

6. The method of claim 1, wherein the target domain knowledge graph is a medical domain knowledge graph.

7. The method of claim 6, further comprising, prior to the determining a target entity node in which the historical search request and the historical recalled advertisement are associated in a target domain knowledge graph and a target attribute node connected to the target entity node:

8. The method of claim 7, wherein constructing the medical knowledge-graph with entity nodes as cores comprises:

9. The method of claim 6, wherein said determining a relevance between said historical search request and said historical recalled ad based on said target attribute node comprises:

10. An advertisement recall apparatus, comprising:

11. An electronic device, comprising:

at least one processor; and

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the advertisement recall method of any of claims 1-9.

12. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the advertisement recall method of any of claims 1-9.