CN111353106A

CN111353106A - Recommendation method and device, electronic equipment and storage medium

Info

Publication number: CN111353106A
Application number: CN202010121352.8A
Authority: CN
Inventors: 邢可
Original assignee: Beike Technology Co Ltd
Current assignee: Seashell Housing Beijing Technology Co Ltd
Priority date: 2020-02-26
Filing date: 2020-02-26
Publication date: 2020-06-30
Anticipated expiration: 2040-02-26
Also published as: CN111353106B

Abstract

The embodiment of the disclosure discloses a recommendation method and device, an electronic device and a storage medium, wherein the recommendation method comprises the following steps: receiving a recommendation request including reference location related information; acquiring initial position related information matched with the reference position related information in the first knowledge graph; generating feature vectors of the matched starting position related information through a neural network, and generating feature vectors of all target areas in the first knowledge graph through the neural network; respectively obtaining the correlation between the feature vector of the matched initial position related information and the feature vector of each target area; and determining at least one target area as a recommendation object based on the correlation between the feature vector of the matched starting position related information and the feature vector of each target area. The method and the device for recommending the target area can recommend the corresponding target area according to the relevant information of the initial position, improve the recommendation accuracy and improve the efficiency of searching the target area.

Description

Recommendation method and device, electronic equipment and storage medium

Technical Field

The present disclosure relates to internet technologies, and in particular, to a recommendation method and apparatus, an electronic device, and a storage medium.

Background

In actual life, a user often needs to rent a house due to the influence of various factors such as living needs, work changes, house buying and selling and the like. Research shows that the quantity of people who need renting houses is millions of times each year due to the influence of work change factors (such as work change, due life and employment and the like). Users rent rooms need to consider many factors, such as rent, commute, community environment, surrounding public facilities, etc.

The current house renting mode is that a user explains house renting requirements to a house intermediary, the house intermediary recommends a plurality of house sources with help, and then the user selects whether a proper house source exists according to the conditions (rent, commute, community environment, peripheral public facilities and the like) of each house source; or browsing the house source information one by one on house renting software or house renting websites to find a proper house source. The existing house renting mode needs to consume a large amount of energy, a satisfied house source is searched for a long time, and house renting efficiency is low.

Disclosure of Invention

The embodiment of the disclosure provides a recommendation method and device, electronic equipment and a storage medium, which can be used for house renting recommendation to improve house renting efficiency.

In one aspect of the disclosed embodiments, a recommendation method is provided, including:

receiving a recommendation request, wherein the recommendation request comprises reference position related information;

acquiring initial position related information matched with the reference position related information in the first knowledge graph; the first knowledge graph is constructed on the basis of the initial position related information, the target area, the relationship between the initial position related information and the target area, and the attribute value of the target area in the data source;

generating a feature vector of the matched start position related information through a neural network, and generating a feature vector of each target region in the first knowledge graph through the neural network; the neural network is obtained by performing feature learning on a first knowledge graph in advance;

respectively obtaining the correlation between the feature vector of the matched initial position related information and the feature vector of each target area;

and determining at least one target area as a recommendation object based on the correlation between the feature vector of the matched starting position related information and the feature vector of each target area.

Optionally, in the method according to any of the above embodiments of the present disclosure, the start position related information includes any one or more of: company identification ID, office building ID, office park ID, location information, regional information; the target area includes: a residential area ID;

the attributes of the target area include any one or more of: rent, community environment, community age, property management, house source quality, peripheral facilities, commuting convenience and shopping convenience.

Optionally, in the method according to any of the above embodiments of the present disclosure, the start position related information includes any one or more of: residential area ID, location information, area information; the target area includes: office building ID, office park ID, regional information; the attributes of the target area include any one or more of: regional environment, property management, parking convenience, parking fee, dining convenience, commuting convenience and shopping convenience.

Optionally, in the method according to any one of the above embodiments of the present disclosure, after determining at least one of the target regions as a recommendation object, the method further includes:

acquiring the attribute and the attribute value of the recommended object from a database or the first knowledge graph;

generating and outputting a recommended object list; wherein the recommended object list includes related information of at least one recommended object, and the related information of the at least one recommended object includes: a target area ID of the recommended object, and at least one attribute and attribute value of the recommended object.

Optionally, in the method according to any of the above embodiments of the present disclosure, the operation of constructing the first knowledge graph based on the starting position related information, the target area, the relationship between the starting position related information and the target area, and the attribute value of the target area in the data source includes:

collecting metadata from a database within a specified range, the metadata comprising: user information related to the initial position related information, target area information corresponding to each user information, and attributes and attribute values of each target area;

determining the relation between the related information of each initial position and each target area in the metadata;

obtaining the data source based on the metadata, the relation between the relevant information of each initial position in the metadata and each target area;

and constructing the first knowledge graph based on a data source consisting of the metadata and the relation between each starting position related information in the metadata and each target area.

Optionally, in the method according to any one of the above embodiments of the present disclosure, the determining a relationship between each piece of start position related information in the metadata and each piece of target area includes:

and if the number of users corresponding to the same target area by the same initial position related information is greater than a first preset value, or the ratio of the number of users corresponding to the same target area by the same initial position related information to the total number of users corresponding to all target areas by the same initial position related information is greater than a second preset value, determining that the relationship between the initial position related information and the target areas in the metadata is favorite.

Optionally, in the method according to any one of the above embodiments of the present disclosure, after the collecting metadata in a specified range from the database, the method further includes:

pre-processing the metadata, the pre-processing comprising any one or more of: removing duplication of target area information corresponding to user information, and performing entity alignment on nodes in the metadata, wherein the nodes comprise: a starting position related information node, a target area node and an attribute value node.

Optionally, in the method according to any one of the above embodiments of the present disclosure, the generating, by a neural network, the feature vector of the matched start position-related information, and the generating, by the neural network, the feature vector of each target region in the first knowledge graph includes:

segmenting the first knowledge-graph into a second knowledge-graph and a plurality of third knowledge-graphs respectively aiming at each attribute of the target area; wherein the second knowledge-graph comprises: the starting position related information, the target area, the relation between the starting position related information and the target area; the third knowledge-graph for one attribute of a target region comprises: a target region, the one attribute, and an attribute value of the one attribute;

respectively generating a node sequence of the second knowledge graph and a node sequence of each third knowledge graph through a random walk strategy with bias execution; wherein each of the node sequences includes a plurality of nodes in a certain order in the corresponding second or third knowledge-graph, the nodes in the second knowledge-graph include a start position-related information node for representing start position-related information and a target region node for representing a target region, and the nodes in the third knowledge-graph include a target region node for representing a target region and an attribute node for representing an attribute;

inputting the node sequence of the second knowledge graph into the neural network, and obtaining the feature vector of each initial position related information in the second knowledge graph through the neural network, wherein the feature vector of each initial position related information in the second knowledge graph comprises the matched feature vector of the initial position related information; respectively inputting the node sequences of the third knowledge graphs into the neural network, and obtaining the feature vectors of the target areas in the third knowledge graphs through the neural network;

and respectively aiming at each target region, obtaining a fusion feature vector of each attribute fused with the target region based on the feature vectors of the target regions in the third knowledge maps.

Optionally, in the method according to any one of the above embodiments of the present disclosure, the obtaining the correlation between the feature vector of the matched start position related information and the feature vector of each target region respectively includes:

and respectively acquiring the correlation between the feature vector of the matched initial position related information and the fusion feature vector of each target area.

Optionally, in the method according to any of the above embodiments of the present disclosure, the obtaining of the neural network based on feature learning of the first knowledge graph includes:

segmenting the first knowledge-graph into a second knowledge-graph and a plurality of third knowledge-graphs respectively aiming at each attribute of the target area;

respectively generating a node sequence of the second knowledge graph and a node sequence of each third knowledge graph through the random walk strategy with the bias execution;

inputting the node sequence of the second knowledge graph into an initial neural network, and obtaining the characteristic vector of the related information of each initial position in the second knowledge graph through the initial neural network; respectively inputting the node sequences of the third knowledge graphs into the initial neural network, and obtaining the feature vectors of the target areas in the third knowledge graphs through the initial neural network;

obtaining a log-likelihood function value between each node and a context node of each node in the node sequence in each node sequence;

adjusting parameters of the initial neural network to maximize a sum of log-likelihood function values between each node and the context node of each node in all the node sequences.

In another aspect of the disclosed embodiments, there is provided a recommendation apparatus, including:

the device comprises a receiving module, a recommending module and a recommending module, wherein the recommending module is used for receiving a recommending request which comprises the related information of the reference position;

the first acquisition module is used for acquiring the relevant information of the initial position matched with the relevant information of the reference position in the first knowledge graph; the first knowledge graph is constructed on the basis of the initial position related information, the target area, the relationship between the initial position related information and the target area, and the attribute value of the target area in the data source;

the first generation module is used for generating a feature vector of the matched starting position related information through a neural network and generating a feature vector of each target area in the first knowledge graph through the neural network; the neural network is obtained by performing feature learning on a first knowledge graph in advance;

a second obtaining module, configured to obtain correlations between the feature vectors of the matched start position related information and the feature vectors of each target region, respectively;

and the determining module is used for determining at least one target area as a recommendation object based on the correlation between the feature vector of the matched starting position related information and the feature vector of each target area.

Optionally, in an apparatus according to any of the above embodiments of the present disclosure, the start position related information includes any one or more of: company identification ID, office building ID, office park ID, location information, regional information; the target area includes: a residential area ID;

Optionally, in an apparatus according to any of the above embodiments of the present disclosure, the start position related information includes any one or more of: residential area ID, location information, area information; the target area includes: office building ID, office park ID, regional information; the attributes of the target area include any one or more of: regional environment, property management, parking convenience, parking fee, dining convenience, commuting convenience and shopping convenience.

Optionally, in an apparatus according to any of the above embodiments of the present disclosure, the apparatus further includes:

the third acquisition module is used for acquiring the attribute and the attribute value of the recommended object from a database or the first knowledge graph;

the second generation module is used for generating and outputting a recommended object list; wherein the recommended object list includes related information of at least one recommended object, and the related information of the at least one recommended object includes: a target area ID of the recommended object, and at least one attribute and attribute value of the recommended object.

Optionally, in an apparatus according to any of the above embodiments of the present disclosure, the apparatus further includes: the construction module is used for constructing and obtaining the first knowledge graph based on the initial position related information, the target area, the relationship between the initial position related information and the target area, and the attribute value of the target area in the data source; the building module comprises:

the collection unit is used for collecting metadata in a specified range from a database, and the metadata comprises: user information related to the initial position related information, target area information corresponding to each user information, and attributes and attribute values of each target area;

a determining unit, configured to determine a relationship between each start position related information in the metadata and each target area;

the first acquisition unit is used for acquiring the data source based on the metadata, the relation among the initial position related information and the target areas in the metadata;

and the construction unit is used for constructing the first knowledge graph based on a data source which is composed of the metadata and the relation between each initial position related information and each target area in the metadata.

Optionally, in an apparatus according to any of the above embodiments of the present disclosure, the determining unit is specifically configured to:

Optionally, in an apparatus according to any of the above embodiments of the present disclosure, the building module further includes:

a preprocessing unit, configured to perform preprocessing on the metadata, where the preprocessing includes any one or more of: removing duplication of target area information corresponding to user information, and performing entity alignment on nodes in the metadata, wherein the nodes comprise: a starting position related information node, a target area node and an attribute value node.

Optionally, in an apparatus according to any one of the above embodiments of the present disclosure, the first obtaining module includes:

a dividing unit configured to divide the first knowledge graph into a second knowledge graph and a plurality of third knowledge graphs for respective attributes of a target region; wherein the second knowledge-graph comprises: the starting position related information, the target area, the relation between the starting position related information and the target area; the third knowledge-graph for one attribute of a target region comprises: a target region, the one attribute, and an attribute value of the one attribute;

a generating unit, configured to generate a node sequence of the second knowledge graph and a node sequence of each third knowledge graph respectively through a biased random walk strategy; wherein each of the node sequences includes a plurality of nodes in a certain order in the corresponding second or third knowledge-graph, the nodes in the second knowledge-graph include a start position-related information node for representing start position-related information and a target region node for representing a target region, and the nodes in the third knowledge-graph include a target region node for representing a target region and an attribute node for representing an attribute;

a second obtaining unit, configured to input the node sequence of the second knowledge graph into the neural network, and obtain, through the neural network, a feature vector of each initial position related information in the second knowledge graph, where the feature vector of each initial position related information in the second knowledge graph includes the feature vector of the matched initial position related information; respectively inputting the node sequences of the third knowledge graphs into the neural network, and obtaining the feature vectors of the target areas in the third knowledge graphs through the neural network;

and the fusion unit is used for obtaining fusion feature vectors of the target regions for fusing the attributes based on the feature vectors of the target regions in the third knowledge maps respectively aiming at the target regions.

Optionally, in an apparatus according to any of the above embodiments of the present disclosure, the second obtaining module is specifically configured to:

Optionally, in the apparatus according to any one of the above embodiments of the present disclosure, the second obtaining unit is further configured to input the node sequence of the second knowledge graph into an initial neural network, and obtain, through the initial neural network, a feature vector of information related to each starting position in the second knowledge graph; respectively inputting the node sequences of the third knowledge graphs into the initial neural network, and obtaining the feature vectors of the target areas in the third knowledge graphs through the initial neural network;

the device further comprises:

the training module is used for acquiring a log-likelihood function value between each node and a context node of each node in the node sequence; adjusting parameters of the initial neural network to maximize a sum of log-likelihood function values between each node and the context node of each node in all the node sequences, thereby obtaining the neural network.

In another aspect of the disclosed embodiments, an electronic device is provided, including:

a memory for storing a computer program;

a processor for executing the computer program stored in the memory, and the computer program, when executed, implements the method of any of the above embodiments of the present disclosure.

In yet another aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, and the computer program, when executed by a processor, implements the method according to any of the above embodiments of the present disclosure.

Based on the recommendation method and apparatus, the electronic device, and the storage medium provided by the above embodiments of the present disclosure, a first knowledge graph may be constructed in advance based on the start position related information, the target region, the relationship between the start position related information and the target region, and the attribute value of the target region in the data source, a neural network may be obtained based on feature learning performed on the first knowledge graph, after receiving a recommendation request, first obtaining start position related information in the first knowledge graph that is matched with reference position related information in the recommendation request, then generating a feature vector of the matched start position related information and feature vectors of each target region in the first knowledge graph through the neural network, and then obtaining correlations between the feature vector of the matched start position related information and the feature vectors of each target region respectively, furthermore, based on the correlation between the feature vectors of the matched start position related information and the feature vectors of the target regions, at least one target region is determined to be used as a recommendation object to feed back to a user, so that the corresponding target region can be recommended for the start position related information, the recommendation accuracy can be improved, and the efficiency of searching for the target region can be improved.

In an application of the embodiment of the present disclosure, the method and the system can be used for recommending a proper living area for a user by integrating various conditions (i.e., attribute values of various attributes, such as rent, community environment, community year, property management, house source quality, peripheral facilities, commuting convenience, shopping convenience, and the like) of each living area (i.e., a target area, such as a living area ID and the like) according to information related to a working area of the user (i.e., starting location related information, such as a company ID, an office building ID, an office park ID, location information, area information, and the like), so that a recommendation accuracy is improved, and efficiency of renting or buying a house by the user can be improved.

In another application of the embodiment of the present disclosure, the method and the device can be used for recommending a proper work area for a user by integrating various conditions (i.e., attribute values of various attributes, such as area environment, property management, parking convenience, parking fee, dining convenience, commuting convenience, shopping convenience, and the like) of each work area (i.e., a target area, such as an office building ID, an office park ID, area information, and the like) according to the living area related information (i.e., starting position related information, such as a living area ID, position information, area information, and the like) of the user, so that the recommendation accuracy is improved, and convenience in work, passage, and the like of the user can be improved.

The technical solution of the present disclosure is further described in detail by the accompanying drawings and examples.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the disclosure and together with the description, serve to explain the principles of the disclosure.

The present disclosure may be more clearly understood from the following detailed description, taken with reference to the accompanying drawings, in which:

FIG. 1 is a flow chart of one embodiment of a recommendation method of the present disclosure.

FIG. 2 is a flow chart of another embodiment of a recommendation method of the present disclosure.

FIG. 3 is a flow diagram of one embodiment of constructing a first knowledge-graph in an embodiment of the present disclosure.

FIG. 4 is a schematic diagram of a first knowledge-graph in an embodiment of the present disclosure.

Fig. 5 is a flowchart of an embodiment of generating a feature vector of matched start position related information and a feature vector of each target area in the embodiment of the present disclosure.

FIG. 6a is a schematic diagram of a second knowledge-graph and a third knowledge-graph in an embodiment of the disclosure.

FIG. 6b is a schematic diagram of a second knowledge-graph and a third knowledge-graph in an embodiment of the disclosure.

Fig. 7 is a diagram of an example of biased random walk strategy in an embodiment of the present disclosure.

FIG. 8 is a flow diagram of one embodiment of learning features of a first knowledge-graph to obtain a neural network in accordance with an embodiment of the present disclosure.

Fig. 9 is a schematic diagram of a neural network processing node sequence in an embodiment of the disclosure.

Fig. 10 is a schematic structural diagram of an embodiment of the recommendation device of the present disclosure.

Fig. 11 is a schematic structural diagram of another embodiment of the recommendation device of the present disclosure.

Fig. 12 is a schematic structural diagram of an embodiment of an electronic device according to the present disclosure.

Detailed Description

Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that: the relative arrangement of the components and steps, the numerical expressions, and numerical values set forth in these embodiments do not limit the scope of the present disclosure unless specifically stated otherwise.

It will be understood by those of skill in the art that the terms "first," "second," and the like in the embodiments of the present disclosure are used merely to distinguish one element from another, and are not intended to imply any particular technical meaning, nor is the necessary logical order between them.

It is also understood that in embodiments of the present disclosure, "a plurality" may refer to two or more and "at least one" may refer to one, two or more.

It is also to be understood that any reference to any component, data, or structure in the embodiments of the disclosure, may be generally understood as one or more, unless explicitly defined otherwise or stated otherwise.

In addition, the term "and/or" in the present disclosure is only one kind of association relationship describing an associated object, and means that three kinds of relationships may exist, for example, a and/or B may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" in the present disclosure generally indicates that the former and latter associated objects are in an "or" relationship.

It should also be understood that the description of the various embodiments of the present disclosure emphasizes the differences between the various embodiments, and the same or similar parts may be referred to each other, so that the descriptions thereof are omitted for brevity.

Meanwhile, it should be understood that the sizes of the respective portions shown in the drawings are not drawn in an actual proportional relationship for the convenience of description.

The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the disclosure, its application, or uses.

Techniques, methods, and apparatus known to those of ordinary skill in the relevant art may not be discussed in detail but are intended to be part of the specification where appropriate.

It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, further discussion thereof is not required in subsequent figures.

The disclosed embodiments may be applied to electronic devices such as terminal devices, computer systems, servers, etc., which are operational with numerous other general purpose or special purpose computing system environments or configurations. Examples of well known terminal devices, computing systems, environments, and/or configurations that may be suitable for use with electronic devices, such as terminal devices, computer systems, servers, and the like, include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, hand-held or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, networked personal computers, minicomputer systems, mainframe computer systems, distributed cloud computing environments that include any of the above, and the like.

Electronic devices such as terminal devices, computer systems, servers, etc. may be described in the general context of computer system-executable instructions, such as program modules, being executed by a computer system. Generally, program modules may include routines, programs, objects, components, logic, data structures, etc. that perform particular tasks or implement particular abstract data types. The computer system/server may be practiced in distributed cloud computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed cloud computing environment, program modules may be located in both local and remote computer system storage media including memory storage devices.

FIG. 1 is a flow chart of one embodiment of a recommendation method of the present disclosure. As shown in fig. 1, the recommendation method of this embodiment includes:

102, receiving a recommendation request, wherein the recommendation request comprises the relevant information of the reference position.

And 104, acquiring the starting position related information matched with the reference position related information in the first knowledge-graph.

The first knowledge graph is constructed on the basis of the initial position related information, the target area, the relationship between the initial position related information and the target area, and the attribute value of the target area in the data source.

And 106, generating a feature vector of the matched starting position related information through a neural network, and generating a feature vector of each target area in the first knowledge graph through the neural network.

Wherein the neural network is obtained in advance based on feature learning of the first knowledge graph.

And 108, respectively acquiring the correlation between the feature vector of the matched start position related information and the feature vector of each target area.

And 110, determining at least one target area as a recommendation object based on the correlation between the feature vector of the matched start position related information and the feature vector of each target area.

The knowledge graph is an auxiliary information emerging in recent years, and the basic structure of the knowledge graph is a directed heteromorphic graph. The knowledge graph is formally proposed by Google in 2012, 5 month and 17 day, and is a heterogeneous semantic network which reveals the relationship between entities, and can formally describe the real world things and their interrelations. In the knowledge graph, node E ═ { E ═ E₁，e₂，e₃，...，e_|E|Represents an entity (also called a concept, corresponding to the semantics of things), the edge R ═ R₁，r₂，r₃，...，r_|R|Represents various semantic relationships (corresponding to relationships between things) between entities/concepts. A triplet (h, r, t) represents a piece of knowledge, and there is some relationship between two entities, where h represents the head node of the knowledge and t represents the tail node. The set of triplets constitutes a knowledge-graph.

Knowledge graph feature learning (KGE) is a sub-field of network feature learning, and obtains a low-dimensional feature vector for learning each entity and relationship in a knowledge graph, so that the high-dimensional property and the heterogeneity of the knowledge graph are reduced, and the original structure or semantic information in the graph is maintained.

Based on the recommendation method provided by the above embodiment of the present disclosure, a first knowledge graph is constructed in advance based on the initial position related information, the target region, the relationship between the initial position related information and the target region in the data source, the attribute and the attribute value of the target region to represent various semantic relationships between each initial position related information and the target region in the data source and semantic relationships between each target region and each attribute value, and a neural network is obtained based on feature learning of the first knowledge graph, when recommending is performed for a recommendation request, the initial position related information in the first knowledge graph, which is matched with the reference position related information in the recommendation request, may be first obtained, and then a feature vector of the matched initial position related information and a feature vector of each target region in the first knowledge graph are generated through the neural network, and then, respectively obtaining the correlation between the feature vector of the matched initial position related information and the feature vector of each target region, and further determining at least one target region as a recommendation object to feed back to a user based on the correlation between the feature vector of the matched initial position related information and the feature vector of each target region, so that the corresponding target region can be recommended for the initial position related information, the recommendation accuracy can be improved, and the efficiency of searching for the target region can be improved.

In some possible implementations of the embodiments of the present disclosure, the start position related information may include, for example, but is not limited to, any one or more of the following: company Identification (ID), office building ID, office park ID, location information, regional information, etc. office regional related information, where the ID may be, for example, a name, number, etc. that uniquely identifies a corresponding company, office building, office park, etc. Correspondingly, the target area may include, for example but not limited to: residential zone related information such as a residential zone ID that uniquely identifies a residential zone, which may be, for example, a community or a residential zone range.

Optionally, in the above implementation, the attribute of the target area may include, but is not limited to, any one or more of the following: rent, community environment, community age, property management, house source quality, surrounding facilities, commuting convenience, shopping convenience, and the like can arbitrarily represent attributes of living area quality and convenience.

In this embodiment, the method and the device can be used for recommending a proper living area for the user by integrating various conditions (i.e., attribute values of various attributes, such as rent, community environment, community year, property management, house source quality, peripheral facilities, commuting convenience, shopping convenience and the like) of each living area (i.e., a target area, such as a living area ID and the like) according to the work area related information (i.e., the starting position related information, such as a company ID, an office building ID, an office park ID, location information, area information and the like) of the user, so that the recommendation accuracy is improved, and the efficiency of renting or buying the house by the user can be improved.

Alternatively, in other possible implementations of embodiments of the present disclosure, the start position related information may include, for example and without limitation, any one or more of the following: residential zone ID, location information, zone information, etc. Correspondingly, the target area may include, for example but not limited to: office building ID, office campus ID, regional information, etc.

Optionally, in this implementation, the attributes of the target area may include, for example, but are not limited to, any one or more of: the attributes of the convenience of the work area can be arbitrarily represented by the regional environment, property management, parking convenience, parking fee, dining convenience, commuting convenience, shopping convenience and the like.

In the embodiment, the method and the device can be used for integrating various conditions (i.e. attribute values of various attributes, such as regional environment, property management, parking convenience, parking fee, dining convenience, commuting convenience, shopping convenience and the like) of each working area (i.e. target area, such as office building ID, office park ID, area information and the like) according to the living area related information (i.e. starting position related information, such as living area ID, position information, area information and the like) of the user to recommend a proper working area for the user, so that the recommendation accuracy is improved, and the convenience in the aspects of work, traffic and the like of the user can be improved.

FIG. 2 is a flow chart of another embodiment of a recommendation method of the present disclosure. As shown in fig. 2, based on the embodiment shown in fig. 1, after the operation 110, the method may further include:

and 112, acquiring the attribute and the attribute value of the recommended object from the database or the first knowledge graph.

114, generating and outputting a recommendation object list.

Wherein, the recommended object list includes related information of at least one recommended object, and the related information of at least one recommended object may include, but is not limited to: a target area ID of the recommended object, and at least one attribute and attribute value of the recommended object.

For example, in an optional example, when the recommendation object is 3 residential area IDs which are community names, the recommendation object list includes the 3 community names, and rent, community environment, community age, property management, house source quality, surrounding facilities, commute convenience, shopping convenience, and specific information thereof corresponding to each community name. As shown in table 1 below, is a specific content example of the recommended object list.

TABLE 1

Optionally, in the recommendation method of the embodiment of the present disclosure, an operation of constructing the first knowledge graph based on the starting position related information, the target region, the relationship between the starting position related information and the target region, and the attribute value of the target region in the data source is further included. As shown in fig. 3, one embodiment of constructing a first knowledge-graph in the embodiments of the present disclosure includes:

metadata within a specified range (e.g., Beijing, Inc.) is collected from a database 202, which may include, for example, but is not limited to: the information of the user related to the start position, the target area information corresponding to each user information, and the attribute and attribute value of each target area, wherein the user information such as user name, user ID, and the like is used for uniquely identifying one user.

The metadata records which users the related information of the initial position respectively relates to, target area information corresponding to each user, and attributes and attribute values of each target area. For example, in a specific example, the metadata records that 120 users in XX park provide living area information and the user information of the 120 users, wherein each user has specific living area, and specific information of rent, community environment, community age, property management, house source quality, surrounding facilities, commuting convenience and shopping convenience of each living area.

And 204, determining the relation between each piece of starting position related information in the metadata and each target area.

In some possible implementation manners of the embodiment of the present disclosure, the relationship between the start position related information and each target area may include: love, irrelevant; alternatively, the relationship between the start position related information and each target area may be divided into a plurality of levels: like, general, not like. Alternatively, the relationship between the start position related information and each target area may also be divided in other manners, and the embodiment of the present disclosure does not limit the specific division of the relationship between the start position related information and each target area.

In an optional example, when the relationship between the related start position information and each target area includes a favorite relationship and a non-favorite relationship, the relationship between the related start position information and the target area in the metadata may be determined to be a favorite relationship when the number of users corresponding to the same target area by the same related start position information is greater than a first preset value, or a ratio between the number of users corresponding to the same target area by the same related start position information and the total number of users corresponding to all target areas by the same related start position information is greater than a second preset value. Otherwise, if the number of users corresponding to the same target area by the same start position related information is not greater than a first preset value, or the ratio of the number of users corresponding to the same target area by the same start position related information to the total number of users corresponding to all target areas by the same start position related information is not greater than a second preset value, it may be determined that the relationship between the start position related information and the target areas in the metadata is irrelevant.

For example, the relationship between the start position-related information and the target area may be determined as a favorite, when the number of users living in the AA garden by the user who works on the XX garden is more than 20, or the ratio of the number of users living in the AA garden by the user who works on the XX garden to the total number of users living in all the living areas by the user who works on the XX garden is more than 10%; otherwise, the relationship between the start position related information and the target area may be determined to be irrelevant.

Based on this embodiment, the relationship between the start position related information and the target area in the metadata may be determined, so as to perform subsequent target area prediction based on the relationship between the start position related information and the target area.

206, obtaining the data source based on the metadata, the relationship between the information related to each starting position in the metadata and each target area.

And 208, constructing a first knowledge graph based on the metadata and a data source formed by the relationship between the starting position related information and the target areas in the metadata.

Based on the embodiment, the metadata in the specified range can be collected from the database and processed to obtain the data source, so that the first knowledge graph is constructed, and the target area can be predicted based on the first knowledge graph.

Optionally, on the basis of the embodiment shown in fig. 3, after the operation 302, the method may further include:

preprocessing the metadata, wherein the preprocessing may include, for example and without limitation, any one or more of the following: the method comprises the steps of carrying out duplication removal on target area information corresponding to user information, and carrying out entity alignment on nodes in metadata, wherein the nodes comprise: a starting position-related information node (i.e., a node corresponding to the starting position-related information in the knowledge-graph), a target region node (i.e., a node corresponding to the target region in the knowledge-graph), and an attribute value node (i.e., a node corresponding to the attribute value in the knowledge-graph).

The duplication removal is performed on the target area information corresponding to the user information, that is, the duplication removal is performed on the target area information corresponding to the same user information providing the initial position related information, for example, in a renting room scene, duplicate tenant information is removed; when the same tenant information provides a plurality of living area information as target areas, only the nearest living area information of the same tenant is reserved as the living area information of the tenant.

The nodes in the metadata are entity aligned, and are used for unifying the same things corresponding to different references into the same reference, for example, "beijing city" and "beijing" refer to the same, and are unified into the same name.

Based on the embodiment, the metadata is preprocessed, so that the simplicity, accuracy and uniqueness of information in the data source can be ensured, the constructed first knowledge graph can reflect the real semantic structure relationship more accurately, and the accuracy of the subsequent recommended target area is improved.

FIG. 4 is a schematic diagram of a first knowledge-graph in an embodiment of the present disclosure. As shown in fig. 4, in a specific application of the embodiment of the present disclosure to a house renting scenario, the start position related information is specifically a company name, which is used to record in which living area the employees of the company rent the house; the rental housing hot area is a living area which is confirmed to be a favorite relationship in the living area (namely the target area); the attributes of the residential zone include: rent, community environment, surrounding facilities, commuting convenience, shopping convenience.

After the first knowledge graph is structured according to the schema shown in fig. 4, the first knowledge graph can represent the nodes and the relationships between the nodes, that is: the relationship between each start position-related information and each target area may represent the relationship between the employees of each company and each living area when applied to a rental room scene.

In some possible implementation manners of the embodiment of the present disclosure, in operation 106, a Node sequence may be generated on the first knowledge graph through a neural network by using a property that nodes with similar structures and the same attribute values may be grouped together by using a Node-to-vector (Node2Vec), and then, the generated Node sequences are respectively input into a Word-to-vector (Word2Vec) neural network, and the Node sequences are respectively mapped to a low-dimensional space, so as to obtain a feature vector representation of each start position related information and each target area Node, and semantic values of nodes with similar structures and the same attribute are more similar.

Fig. 5 is a flowchart of an embodiment of generating a feature vector of matched start position related information and a feature vector of each target area in the embodiment of the present disclosure. As shown in fig. 5, based on the above embodiment, operation 106 includes:

the first knowledge-graph is divided 302 into a second knowledge-graph and a plurality of third knowledge-graphs for each attribute of the target region.

Wherein the second knowledge-graph comprises: the starting position related information, the target area, the relation between the starting position related information and the target area; the third knowledge-graph for one attribute of a target region comprises: a target area, the one attribute, and an attribute value of the one attribute.

As shown in fig. 6a, the first knowledge maps in the renting scene application shown in fig. 4 are divided into second knowledge maps; as shown in fig. 6b, a schematic diagram of a mode diagram of a third knowledge graph with an attribute of a community environment obtained by segmenting the first knowledge graph in the house renting scenario application shown in fig. 4 is shown, and the mode diagrams of the third knowledge graphs with other attributes obtained by segmenting the first knowledge graph in the house renting scenario application shown in fig. 4 are similar to these mode diagrams, and are not described in detail.

And 304, respectively generating a node sequence of the second knowledge graph and a node sequence (node sequence) of each third knowledge graph through a random walk strategy with bias.

Each node sequence includes a plurality of nodes in a certain order in a corresponding second knowledge graph or a corresponding third knowledge graph, the nodes in the second knowledge graph include a start position related information node for representing start position related information and a target area node for representing a target area, and the nodes in the third knowledge graph include a target area node for representing a target area and an attribute node for representing an attribute, that is, the nodes in the embodiments of the present disclosure are concepts/entities in the knowledge graph.

And 306, inputting the node sequence of the second knowledge graph into the neural network, and obtaining the feature vector of the related information of each initial position in the second knowledge graph through the neural network. And respectively inputting the node sequences of the third knowledge graphs into the neural network, and obtaining the feature vector of each target area in each third knowledge graph through the neural network, so that the feature vector representation of each target area in a specific attribute space can be obtained, and the semantic values of the nodes of the target areas with similar structures and the same attributes in the attribute space are more similar.

And the feature vector of each initial position related information in the second knowledge graph comprises the matched feature vector of the initial position related information.

308, respectively aiming at each target area, obtaining a fusion feature vector of each attribute fused with the target area based on the feature vectors of the target areas in the plurality of third knowledge maps.

Corresponding to the embodiment shown in fig. 5, in operation 108, correlations between the feature vectors of the matched start position related information and the fused feature vectors of the target regions may be obtained respectively.

Because the knowledge-graphs of different attributes have different semantic values, for example, in a rental-house scenario, different rental-house hot-areas may have the same rent, but their shopping convenience or commuting convenience may be different, if the entire knowledge-graph (i.e., the first knowledge-graph) is processed, the semantics of different attributes of the target area may be ignored.

In this embodiment, for a plurality of attributes of the target region, the first knowledge graph is divided into a second knowledge graph and a plurality of third knowledge graphs respectively for the attributes of the target region, so as to embody semantic values of the knowledge graphs of the attributes; respectively generating node sequences of the second knowledge graph and the third knowledge graph with the attributes through a biased random walk strategy, and obtaining the characteristic vector of each node sequence through a neural network model, so that the characteristic vector of the related information of each initial position and the characteristic vector representation of each target area under a specific attribute space are obtained, and the semantic values of the nodes with similar structures and the same attributes are more similar; and determining the recommended object based on the correlation between the feature vector of the matched initial position related information and the fusion feature vector of each target region, so that the recommended object is predicted more comprehensively, accurately and reasonably, and the recommendation accuracy is further improved.

In one possible implementation, continuing with the application example of the house renting scenario of fig. 4 and fig. 6a and 6b, the first knowledge graph is segmented, for example, by a database query language and data acquisition protocol (SPARQL) segmentation technique, to obtain the following 6 knowledge graphs: 1 second best modeIdentifying maps and 5 third knowledge maps corresponding to the 5 attributes, and generating corresponding node sequences S ═ { u ═ u { through random walk strategies with bias in each knowledge map (hereinafter, the second knowledge map and each third knowledge map are referred to specifically)₁，u₂…u_nWhere n is an integer greater than 1. For a given source node u, a random walk of fixed length L is simulated, u_jRepresents the j-th node, the initial node u, in the course of walking₀J is an integer greater than or equal to 0. Node u_jGenerated from the probability distribution, i.e.:

in formula (1), D represents the set of edges in the knowledge-graph, π_vxAnd Z is a preset normalization constant. As shown in fig. 7, the random walk traverses the edge (v, x) and stays at node v. By calculating the transition probability pi on the edge (v, x)_vxJudging the next node in the node sequence, wherein the calculation formula is as follows:

π_vx＝α(t，x)*ω_vx(2)

in the formula (2), ω_vxIs the weight of the edge, can be preset, can default to 1 when not set, α (t, x) represents the bias on the edge between the nodes, and the calculation formula is as follows:

in the formula (3), d_txRepresents the minimum number of hops between nodes t and x, d in FIG. 7_txThe value of (A) belongs to (0,1, 2). As shown in FIG. 7, d_txWhen 0, it means that the node x is the node t itself; d_txWhen 1, the node x is denoted as a node x₁Or x₃；d_txWhen 2, the node x is x₂. p and q are two parameters for supervising random walk, and the values are both numbers larger than 0.

Where p is a Return parameter (Return parameter) that controls the probability of returning to the original node. Setting p to a larger value (p > max (q,1)) ensures that nodes that have been visited will not be sampled in subsequent processes (unless the next node has no other neighbor nodes). If p is set to a small value (p < max (q,1)), the random walk sampling will be backed off by one step, which will bring the sampled node closer to the initial node t.

q is an In-out parameter (In-out parameter) that controls the relationship between the Breadth-first Sampling (BFS) and the Depth-first Sampling (DFS). q allows sampling around or away from the current node. If q >1, the node sequence of random walk samples will be close to the initial node t, so that the node sequence of random walk samples is similar to BFS. Conversely, if q <1, the nodes sampled by random walk will gradually get away from the initial node t, so the collected node sequence is similar to DFS. The sampling can be balanced between BFS and DFS by setting the appropriate parameter q. When p is equal to q is equal to 1, the method is equivalent to the traditional deep walking (deep walk).

By setting different p and q, the node sequences with different skewness can be obtained. In the embodiment of the present disclosure, values of p and q may be set according to a specific application scenario, or when a model of a random walk strategy is trained, a grid search method (grid search) may be used to find optimal p and q. The specific values of p and q and the obtaining mode thereof are not limited in the embodiments of the present disclosure.

Based on the embodiment, the node sequences of the second knowledge graph and the node sequences of the third knowledge graphs can be respectively generated through a random walk strategy with bias execution, so that the possible transfer situations between the nodes can be determined.

In one possible implementation manner, in operation 308, the feature vectors of the third knowledge graph of all attributes of each target area may be added to obtain a sum of the feature vectors, and then the sum of the feature vectors is averaged over the number of attributes to obtain a fusion feature vector fusing semantics of all attributes of each living area. For example, for the house renting scenes shown in fig. 4, fig. 6a, and fig. 6b, after the feature vectors of the same house renting hot area node in the third knowledge graph with different attributes are obtained to represent, the feature vectors of 5 attributes are summed up and then averaged by the following formula (4), so that the fused feature vectors in which all attribute semantics are fused in each house renting hot area are obtained, and the semantic information represented by the fused feature vectors is more comprehensive.

In formula (4), each target region of v (attract) contains a fusion feature vector of all attribute semantics; when m is 5, it means that the target area has 5 attributes, for example, the renting scene shown in fig. 4, fig. 6a, and fig. 6b, and the renting hot area has 5 attributes.

In some possible implementations, the feature vector of the start position related information obtained by operation 306 may be represented as:

v(tenant)v_f(5)

in one possible implementation manner, after obtaining the feature vector of each start position related information in the vector space and the fused feature vector of each target region in the same vector space, in operation 108, the correlation score between the feature vector of the matched start position related information and the fused feature vector of each target region may be obtained by calculating the cosine similarity between the feature vector of the matched start position related information and the fused feature vector of each target region as follows:

Rel(attract，tenant)＝sim(v(attract)，v(tenant)) (6)

in formula (6), sim () is the cosine similarity. In addition, the correlation score between the feature vector of the matched start position related information and the fused feature vector of each target region may also be calculated by other manners, for example, cosine distance.

In one possible implementation manner, after obtaining the correlation score between the feature vector of the matched start position related information and the fused feature vector of each target region, in operation 110, a preset number (for example, 5) of target regions may be selected as recommendation objects according to the sequence from high to low of the correlation score. Or, the correlation scores may be further normalized in the following manner, and then a preset number (for example, 5) of target regions are selected as recommendation objects according to the sequence of the normalized correlation scores from high to low: and (3) rounding up one numerical value in the correlation score conventions 1-5 to obtain a prediction score of the corresponding target area:

therefore, the relevance scores can be selected to select the first few target areas most relevant to the matched starting position relevant information as recommendation objects to recommend the recommendation objects to the user. When the method is applied to a house renting scene, the cosine similarity can be utilized to calculate the correlation score between each company as a house renting user and each house renting hot area, and a plurality of house renting hot areas with the highest prediction scores are selected to be recommended to the house renting user.

Optionally, in the recommendation method of the embodiment of the present disclosure, the method further includes an operation of performing feature learning on the first knowledge graph to obtain the neural network. As shown in fig. 8, an embodiment of learning features of the first knowledge graph to obtain the neural network in the embodiment of the present disclosure includes:

the first knowledge-graph is partitioned 402 into a second knowledge-graph and a plurality of third knowledge-graphs, one for each attribute of the target region.

And 404, respectively generating a node sequence of the second knowledge graph and a node sequence of each third knowledge graph through the random walk strategy with bias.

406, inputting the node sequence of the second knowledge graph into the initial neural network, and obtaining the feature vectors of the related information of each initial position in the second knowledge graph through the initial neural network; and respectively inputting the node sequences of the third knowledge graphs into the initial neural network, and obtaining the feature vectors of the target areas in the third knowledge graphs through the initial neural network.

And 408, obtaining log-likelihood function values between each node and the context node of each node in the node sequence in each node sequence.

And 410, adjusting parameters of the initial neural network so as to maximize the sum of log-likelihood function values between each node and the context node of each node in all the node sequences.

Based on the embodiment, log-likelihood function values between each node and the context node in the node sequence are obtained in each node sequence, and the parameters of the initial neural network are adjusted, so that the sum of the log-likelihood function values between each node and the context node in all the node sequences is maximized, and thus, a more accurate and reasonable node sequence can be obtained, and the feature vector of the context node in one node sequence can be better utilized to learn the feature vector of the current node (namely, a target node), so that the next hop node of the current node is accurately predicted, and therefore, the target region node which is most likely to be moved by the related information node of the current initial position can be accurately predicted.

In the following, with reference to a specific example, the explanation is made on the neural network obtained by learning the features of the first knowledge graph, where the nodes in this embodiment are concepts/entities in the knowledge graph, and the method includes: the node of any one of the three types, namely, the initial position related information node, the target area node and the attribute value node.

1, neural network model: corresponding to the initial neural network or the trained neural network

At least one sequence of nodes for each of the knowledge-graphs has been generated by operation 404 for each of a plurality of independent knowledge-graphs (a second knowledge-graph and each third knowledge-graph). Sequence of nodes with one of S ═ { u ═₁，u₂…u_nFor example, the model of Word2Vec is expanded to a graph node sequence representation model based on a graph node sequence representation model constructed by a three-layer neural network, and the network structure of the model is shown in fig. 9.The objective function based on neural network models is typically taken as the log-likelihood function:

in formula (4), N represents the number of nodes in the current sequence of nodes;

representing a target node u_jBy the target node u_jContext nodes of (1).

Each representing a particular context node. In brief, assume that

Is a non-negative feature vector, where each term represents the number of occurrences of a node in the corresponding context, and the feature vector for the f-th context node is a d-dimensional feature vector, which may be represented as v_f∈R^d。

The front and back 2c position nodes of the node in the node sequence can be selected as training samples for illustration: of the neural network model:

an input layer: randomly initialized feature vector sequences containing 2c positions in the node sequences;

projection layer: summing and averaging the 2c characteristic vectors of the input layer to obtain a projection vector;

an output layer: the result is a complete binary tree, which is a Huffman tree constructed by using nodes appearing in the node sequence as leaf nodes and using the times of each node appearing in the node sequence as weight value, (each branch in the Huffman tree can be used as a first two-classification, the process of establishing the Huffman tree is the process of fitting multiple two-classification and multiple classifications)

The left and right nodes of the non-leaf node are divided into positive examples and coded into

The number of leaf nodes of the Huffman tree is N, and each leaf node is the characteristic vector of the node; there are a total of N-1 non-leaf nodes.

Parameter adjustment and optimization of neural network model

Instantiating the feature vector of each context node in formula (1) by each knowledge graph (second knowledge graph, third knowledge graph), defining the following log-likelihood function as an objective function in the generated node sequence of each specific attribute:

in the formula: e represents a node set in each knowledge graph; t is^uRepresenting a sequence of nodes generated starting from node u; n is a radical of_tIs the length of each node sequence t.

To obtain the objective function, parameter learning and optimization are required. In the neural network model, parameter learning is required to make better use of the feature vectors v of the context nodes_fTo learn the feature vector v of the target node_u. In the embodiment of the present disclosure, the parameters of the neural network are learned by maximizing the log-likelihood function value of equation (9). In some optional examples, the number likelihood function may be optimized by using a loss function (softmax) or a hierarchical loss function (hierarchical softmax), and when the hierarchical loss function is used, the problem of large calculation amount caused by the optimization of the softmax may be simplified, so as to improve the parameter learning efficiency.

Given a target node u_jProjection vector of

L(u_j) Representing root node to target node u_jPath length of (2) of

Can be defined by using the hierarchical softmax

In the form:

in equation (10):

to be provided with

Represents the path L (u)_j) And (5) training parameters of the neural network by adopting a gradient ascending method according to the feature vector of the nth non-leaf node. In the training process, all the node sequences are traversed, and in the process, the target node u_jAnd the feature vectors of its context nodes are updated accordingly. After the loss function values are calculated, an error gradient is obtained by back propagation, and the error gradient is used to update parameters in the neural network model. To facilitate updating θ in each step, it is calculated as follows

Gradient (2):

therefore, the temperature of the molten metal is controlled,

the updating can be done by:

where μ denotes the learning rate of the neural network, which can be determined by the performance of the neural network.

Through the above calculation, the target node u_jThe feature vector of the context node of (2) may be updated by:

through the training of the neural network, each node u can be trained_jAnd the feature vectors of the context nodes in the node sequence are gradually adjusted and optimized, and after the neural network training is completed, the feature vectors of each optimized node and the context nodes in the node sequence can be obtained, so that the feature vector of each node can be more accurately and reasonably represented.

Any of the recommendation methods provided by embodiments of the present disclosure may be performed by any suitable device having data processing capabilities, including but not limited to: terminal equipment, a server and the like. Alternatively, any of the recommendation methods provided by the embodiments of the present disclosure may be executed by a processor, such as the processor executing any of the recommendation methods mentioned by the embodiments of the present disclosure by calling a corresponding instruction stored in a memory. And will not be described in detail below.

Those of ordinary skill in the art will understand that: all or part of the steps for implementing the method embodiments may be implemented by hardware related to program instructions, and the program may be stored in a computer readable storage medium, and when executed, the program performs the steps including the method embodiments; and the aforementioned storage medium includes: various media that can store program codes, such as ROM, RAM, magnetic or optical disks.

Fig. 10 is a schematic structural diagram of an embodiment of the recommendation device of the present disclosure. The recommendation device of this embodiment can be used to implement the above recommendation method embodiments of the present disclosure. As shown in fig. 10, the recommendation apparatus of this embodiment includes: the device comprises a receiving module, a first obtaining module, a first generating module, a second obtaining module and a determining module. Wherein:

the device comprises a receiving module and a recommending module, wherein the recommending module is used for receiving a recommending request which comprises the related information of the reference position.

And the first acquisition module is used for acquiring the relevant information of the initial position matched with the relevant information of the reference position in the first knowledge graph. The first knowledge graph is constructed on the basis of the initial position related information, the target area, the relationship between the initial position related information and the target area, and the attribute value of the target area in the data source.

The first generation module is used for generating a feature vector of the matched starting position related information through a neural network and generating a feature vector of each target area in the first knowledge graph through the neural network; wherein the neural network is obtained in advance based on feature learning of the first knowledge graph.

And the second acquisition module is used for respectively acquiring the correlation between the feature vector of the matched initial position related information and the feature vector of each target area.

Based on the recommendation method provided by the above embodiment of the present disclosure, a first knowledge graph may be constructed in advance based on the initial position related information in the data source, the target region, the relationship between the initial position related information and the target region, the attribute and the attribute value of the target region, a neural network may be obtained based on feature learning performed on the first knowledge graph, after receiving the recommendation request, first obtaining the initial position related information in the first knowledge graph, which is matched with the reference position related information in the recommendation request, then generating the feature vector of the matched initial position related information and the feature vector of each target region in the first knowledge graph through the neural network, then obtaining the correlation between the feature vector of the matched initial position related information and the feature vectors of each target region, and further, based on the correlation between the feature vector of the matched initial position related information and the feature vectors of each target region, at least one target area is determined to be used as a recommendation object to be fed back to a user, so that the corresponding target area can be recommended according to the relevant information of the initial position, the recommendation accuracy can be improved, and the efficiency of searching for the target area is improved.

Based on the embodiment, the method and the device can be used for recommending a proper living area for the user by integrating various conditions (namely attribute values of various attributes, such as rent, community environment, community age, property management, house source quality, peripheral facilities, commuting convenience, shopping convenience and the like) of each living area (namely target area, namely living area ID and the like) aiming at the work area related information (namely starting position related information, such as company ID, office building ID, office park ID, position information, area information and the like) of the user, so that the recommendation accuracy is improved, and the house renting or house buying efficiency of the user can be improved.

Based on the embodiment, the method and the device can be used for integrating various conditions (namely attribute values of various attributes, such as regional environment, property management, parking convenience, parking fee, dining convenience, commuting convenience, shopping convenience and the like) of various working areas (namely target areas, such as office building IDs, office park IDs, area information and the like) according to the living area related information (namely starting position related information, such as a living area ID, position information, area information and the like) of the user to recommend the appropriate working area for the user, so that the recommendation accuracy is improved, and the convenience in the aspects of working, passing and the like of the user can be improved.

Fig. 11 is a schematic structural diagram of another embodiment of the recommendation device of the present disclosure. As shown in fig. 11, compared with the embodiment shown in fig. 10, the recommendation device of this embodiment further includes: the device comprises a third acquisition module and a second generation module. Wherein:

and the third acquisition module is used for acquiring the attribute and the attribute value of the recommended object from a database or the first knowledge graph.

And the second generation module is used for generating and outputting the recommended object list. Wherein the recommended object list includes related information of at least one recommended object, and the related information of the at least one recommended object includes: a target area ID of the recommended object, and at least one attribute and attribute value of the recommended object.

In addition, referring to fig. 11 again, in another embodiment of the recommendation device of the present disclosure, the recommendation device of the present disclosure may further include: and the construction module is used for constructing and obtaining the first knowledge graph based on the initial position related information, the target area, the relationship between the initial position related information and the target area, and the attribute value of the target area in the data source.

In some possible implementations, the building module includes: the collection unit is used for collecting metadata in a specified range from a database, and the metadata comprises: user information related to the initial position related information, target area information corresponding to each user information, and attributes and attribute values of each target area; a determining unit, configured to determine a relationship between each start position related information in the metadata and each target area; the first acquisition unit is used for acquiring the data source based on the metadata, the relation among the initial position related information and the target areas in the metadata; and the construction unit is used for constructing the first knowledge graph based on a data source which is composed of the metadata and the relation between each initial position related information and each target area in the metadata.

In some optional examples, the determining unit is specifically configured to: and if the number of users corresponding to the same target area by the same initial position related information is greater than a first preset value, or the ratio of the number of users corresponding to the same target area by the same initial position related information to the total number of users corresponding to all target areas by the same initial position related information is greater than a second preset value, determining that the relationship between the initial position related information and the target areas in the metadata is favorite.

In addition, in other possible implementations, the building module may further include: a preprocessing unit, configured to perform preprocessing on the metadata, where the preprocessing includes any one or more of: removing duplication of target area information corresponding to user information, and performing entity alignment on nodes in the metadata, wherein the nodes comprise: a starting position related information node, a target area node and an attribute value node.

In some possible implementations, the first obtaining module includes: a dividing unit configured to divide the first knowledge graph into a second knowledge graph and a plurality of third knowledge graphs for respective attributes of a target region; wherein the second knowledge-graph comprises: the starting position related information, the target area, the relation between the starting position related information and the target area; the third knowledge-graph for one attribute of a target region comprises: a target region, the one attribute, and an attribute value of the one attribute; a generating unit, configured to generate a node sequence of the second knowledge graph and a node sequence of each third knowledge graph respectively through a biased random walk strategy; wherein each of the node sequences includes a plurality of nodes in a certain order in the corresponding second or third knowledge-graph, the nodes in the second knowledge-graph include a start position-related information node for representing start position-related information and a target region node for representing a target region, and the nodes in the third knowledge-graph include a target region node for representing a target region and an attribute node for representing an attribute; a second obtaining unit, configured to input the node sequence of the second knowledge graph into the neural network, and obtain, through the neural network, a feature vector of each initial position related information in the second knowledge graph, where the feature vector of each initial position related information in the second knowledge graph includes the feature vector of the matched initial position related information; respectively inputting the node sequences of the third knowledge graphs into the neural network, and obtaining the feature vectors of the target areas in the third knowledge graphs through the neural network; and the fusion unit is used for obtaining fusion feature vectors of the target regions for fusing the attributes based on the feature vectors of the target regions in the third knowledge maps respectively aiming at the target regions.

In some possible implementation manners, the second obtaining module is specifically configured to: and respectively acquiring the correlation between the feature vector of the matched initial position related information and the fusion feature vector of each target area.

In addition, in some possible implementation manners, the second obtaining unit is further configured to input the node sequence of the second knowledge graph into an initial neural network, and obtain, through the initial neural network, a feature vector of information related to each starting position in the second knowledge graph; and respectively inputting the node sequences of the third knowledge graphs into the initial neural network, and obtaining the feature vectors of the target areas in the third knowledge graphs through the initial neural network. Accordingly, referring again to fig. 11, in yet another embodiment of the recommendation device of the present disclosure, the recommendation device of the present disclosure may further include: the training module is used for acquiring a log-likelihood function value between each node and a context node of each node in the node sequence; adjusting parameters of the initial neural network to maximize a sum of log-likelihood function values between each node and the context node of each node in all the node sequences, thereby obtaining the neural network.

In addition, an embodiment of the present disclosure also provides an electronic device, including:

a memory for storing a computer program;

a processor, configured to execute the computer program stored in the memory, and when the computer program is executed, implement the recommendation method according to any of the above embodiments of the present disclosure.

Next, an electronic apparatus according to an embodiment of the present disclosure is described with reference to fig. 12. The electronic device may be either or both of the first device and the second device, or a stand-alone device separate from them, which stand-alone device may communicate with the first device and the second device to receive the acquired input signals therefrom.

Fig. 12 is a schematic structural diagram of an embodiment of an electronic device according to the present disclosure. As shown in fig. 12, the electronic device includes one or more processors and memory.

The processor may be a Central Processing Unit (CPU) or other form of processing unit having data processing capabilities and/or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

The memory may include one or more computer program products that may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. The volatile memory may include, for example, Random Access Memory (RAM), cache memory (cache), and/or the like. The non-volatile memory may include, for example, Read Only Memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium and executed by a processor to implement the recommendation methods of the various embodiments of the disclosure described above and/or other desired functionality.

In one example, the electronic device may further include: an input device and an output device, which are interconnected by a bus system and/or other form of connection mechanism (not shown).

The input device may also include, for example, a keyboard, a mouse, and the like.

The output device may output various information including the determined distance information, direction information, and the like to the outside. The output devices may include, for example, a display, speakers, a printer, and a communication network and remote output devices connected thereto, among others.

Of course, for simplicity, only some of the components of the electronic device relevant to the present disclosure are shown in fig. 12, omitting components such as buses, input/output interfaces, and the like. In addition, the electronic device may include any other suitable components, depending on the particular application.

In addition to the above-described methods and apparatus, embodiments of the present disclosure may also be a computer program product comprising computer program instructions that, when executed by a processor, cause the processor to perform the steps in the recommendation method according to various embodiments of the present disclosure described in the above-mentioned part of the specification.

The computer program product may write program code for carrying out operations for embodiments of the present disclosure in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server.

Furthermore, embodiments of the present disclosure provide a computer-readable storage medium having stored thereon computer program instructions which, when executed by a processor, cause the processor to perform the steps in the recommendation method according to various embodiments of the present disclosure described in the above section of this specification.

The computer-readable storage medium may take any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may include, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The foregoing describes the general principles of the present disclosure in conjunction with specific embodiments, however, it is noted that the advantages, effects, etc. mentioned in the present disclosure are merely examples and are not limiting, and they should not be considered essential to the various embodiments of the present disclosure. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the disclosure is not intended to be limited to the specific details so described.

In the present specification, the embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same or similar parts in the embodiments are referred to each other. For the system embodiment, since it basically corresponds to the method embodiment, the description is relatively simple, and for the relevant points, reference may be made to the partial description of the method embodiment.

The block diagrams of devices, apparatuses, systems referred to in this disclosure are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".

The methods and apparatus of the present disclosure may be implemented in a number of ways. For example, the methods and apparatus of the present disclosure may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order for the steps of the method is for illustration only, and the steps of the method of the present disclosure are not limited to the order specifically described above unless specifically stated otherwise. Further, in some embodiments, the present disclosure may also be embodied as programs recorded in a recording medium, the programs including machine-readable instructions for implementing the methods according to the present disclosure. Thus, the present disclosure also covers a recording medium storing a program for executing the method according to the present disclosure.

It is also noted that in the devices, apparatuses, and methods of the present disclosure, each component or step can be decomposed and/or recombined. These decompositions and/or recombinations are to be considered equivalents of the present disclosure.

The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the disclosure. Thus, the present disclosure is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit embodiments of the disclosure to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.

Claims

1. A recommendation method, comprising:

2. The method of claim 1, wherein the starting position related information comprises any one or more of: company identification ID, office building ID, office park ID, location information, regional information; the target area includes: a residential area ID;

3. The method of claim 1, wherein the starting position related information comprises any one or more of: residential area ID, location information, area information; the target area includes: office building ID, office park ID, regional information; the attributes of the target area include any one or more of: regional environment, property management, parking convenience, parking fee, dining convenience, commuting convenience and shopping convenience.

4. The method according to any one of claims 1-3, wherein after determining at least one of the target regions as a recommendation object, further comprising:

5. The method according to any one of claims 1 to 4, wherein the operation of constructing the first knowledge graph based on the start position related information, the target region, the relationship between the start position related information and the target region, and the attribute value of the target region in the data source comprises:

6. The method of claim 5, wherein determining the relationship between each start position related information in the metadata and each target area comprises:

7. The method according to claim 5 or 6, wherein after collecting the metadata in the specified range from the database, further comprising:

8. A recommendation device, comprising:

9. An electronic device, comprising:

a memory for storing a computer program;

a processor for executing a computer program stored in the memory, and when executed, implementing the method of any of the preceding claims 1-7.

10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the method of any one of the preceding claims 1 to 7.