CN111738807A

CN111738807A - Method, computing device, and computer storage medium for recommending target objects

Info

Publication number: CN111738807A
Application number: CN202010715311.1A
Authority: CN
Inventors: 胡强; 乌景猛
Original assignee: Shanghai Zhongdan Information Technology Co ltd
Current assignee: Shanghai Zhongdan Information Technology Co ltd
Priority date: 2020-07-23
Filing date: 2020-07-23
Publication date: 2020-10-02
Anticipated expiration: 2040-07-23
Also published as: CN111738807B

Abstract

The present disclosure relates to a method, computing device, and computer storage medium for recommending a target object. The method comprises the following steps: extracting image features of an image about a target object; generating a tag feature associated with the comment information based on the comment information; counting predetermined operations on the target object within a predetermined time interval to generate operation characteristics; fusing the image features, the label features and the operation features to generate description features; determining the similarity between the current target object aimed by the user operation and the target object to be selected based on the description characteristics, and determining a display sequence for displaying the target object to be recommended to the user based on the similarity; and displaying the images of the target objects to be recommended based on the display sequence. The method and the device can quickly enable the recommended target object to accurately match the preference of the user about the current target object.

Description

Method, computing device, and computer storage medium for recommending target objects

Technical Field

The present disclosure relates generally to machine learning, and in particular, to methods, computing devices, and computer storage media for recommending target objects.

Background

The conventional scheme for recommending a target object is, for example: the recommendation of the product or the content is performed based on the operation characteristics such as the click rate for the user history, or the similar product is searched for the recommendation based on the name of the product purchased by the user history. However, in the above-described conventional scheme of recommending a target object, there is a large gap between a recommended commodity and a commodity actually preferred by a user, and the speed of determining the recommended commodity is slow, it is difficult to quickly and accurately match the user's preference for the current commodity so as to quickly and accurately recommend an alternative commodity.

In summary, it is difficult for the conventional scheme of recommending target objects to quickly make the recommended target objects accurately match the user's preferences with respect to the current goods.

Disclosure of Invention

The present disclosure provides a method, a computing device, and a computer storage medium for recommending a target object, which enable quickly making a recommended target object accurately match a user's preference with respect to a current target object.

According to a first aspect of the present disclosure, a method for recommending a target object is provided. The method comprises the following steps: extracting image features of an image about a target object; obtaining comment information about the target object so as to generate a tag feature associated with the comment information based on the comment information; counting predetermined operations on the target object within a predetermined time interval to generate operation characteristics; fusing the image feature, the tag feature and the operation feature to generate a description feature about the target object, the dimension of the description feature being smaller than the sum of the dimensions of the image feature, the tag feature and the operation feature; determining the similarity between the current target object aimed by the user operation and the target object to be selected based on the description characteristics, and determining a display sequence for displaying the target object to be recommended to the user based on the similarity; and displaying the images of the target objects to be recommended based on the display sequence.

According to a second aspect of the present invention, there is also provided a computing device comprising: at least one processing unit; at least one memory coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions when executed by the at least one processing unit, cause the apparatus to perform the method of the first aspect of the disclosure.

According to a third aspect of the present disclosure, there is also provided a computer-readable storage medium. The computer readable storage medium has stored thereon a computer program which, when executed by a machine, performs the method of the first aspect of the disclosure.

In some embodiments, fusing the image features, the tag features, and the operational features to generate descriptive features about the target object comprises: stitching the image features, the label features and the operation features to generate input features for inputting the first neural network model; and determining descriptive features for the target object via a first neural network model based on the input features, the first neural network model being trained via a plurality of samples.

In some embodiments, the first neural network model is constructed based on a self-coding network.

In some embodiments, generating the tag feature associated with the review information based on the review information comprises: converting comment information associated with the target object into a feature vector; determining attribute probabilities about the review information based on the feature vectors, the attributes being associated with the type of the target object; determining attributes about the review information based on the determined attribute probabilities; determining, based on the review information, an emotional tendency probability for the attribute for determining an emotional tendency for the attribute; determining comment information to be selected based on the determined emotional tendency and comment information about the attributes; and extracting keyword information associated with the emotional tendency of the attribute in the comment information to be selected so as to generate a label characteristic associated with the target object based on the attribute and the keyword information.

In some embodiments, the attribute includes at least one of smell, taste, price, color, size, function, quality, interior, electricity consumption, material, price, and the target object is a commodity.

In some embodiments, determining the candidate review information based on the determined emotional propensity for the attribute and the review information comprises: in response to determining that the emotional tendency of the attribute meets a predetermined condition, determining whether the length of the comment information is greater than or equal to a predetermined length threshold; and in response to determining that the length of the comment information is greater than or equal to a predetermined length threshold, determining the comment information as the comment information to be selected.

In some embodiments, based on the description features, determining a similarity between the current target object for the user operation and the target object to be selected, for determining a display order for displaying the target object to be recommended to the user based on the similarity includes: obtaining the description characteristics of a current target object aimed at by user operation; performing locality sensitive hash calculation aiming at the description characteristics of the current target object to generate a first hash value; obtaining the description characteristics of a target object to be selected, wherein the packet hash value is the same as the first hash value, and the packet hash value is generated by performing local sensitive hash calculation aiming at the description characteristics of the target object; and respectively calculating the similarity of the description features of the current target object and the description features of the target object to be selected so as to determine the sequence of displaying the target objects to the user based on the similarity calculation result.

In some embodiments, the method for recommending a target object further comprises: performing locality sensitive hash calculation on the description characteristics of each target object in all target objects to generate a packet hash value of each target object; and associating multiple target objects having the same packet hash value with the same hash bucket.

In some embodiments, the predetermined operation on the target object comprises at least one of: clicking operation, adding a shopping cart operation, ordering operation and setting as attention aiming at the target object. This summary is provided to introduce a selection of concepts in a simplified form that are further described below in the detailed description. This summary is not intended to identify key features or essential features of the disclosure, nor is it intended to be used to limit the scope of the disclosure.

Drawings

Fig. 1 shows a schematic diagram of a system for implementing a method for recommending a target object according to an embodiment of the present disclosure.

Fig. 2 shows a flow diagram of a method for recommending a target object according to an embodiment of the present disclosure.

Fig. 3 shows a schematic diagram of a first neural network model for generating descriptive features about a target object, in accordance with an embodiment of the present disclosure.

Fig. 4 shows a flow diagram of a method for generating tag features in accordance with an embodiment of the present disclosure.

Fig. 5 shows a flowchart of a method for determining a display order of target objects to be recommended according to an embodiment of the present disclosure.

FIG. 6 schematically shows a block diagram of an electronic device suitable for use to implement an embodiment of the disclosure.

Like or corresponding reference characters designate like or corresponding parts throughout the several views.

Detailed Description

Preferred embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While the preferred embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

The term "include" and variations thereof as used herein is meant to be inclusive in an open-ended manner, i.e., "including but not limited to". Unless specifically stated otherwise, the term "or" means "and/or". The term "based on" means "based at least in part on". The terms "one example embodiment" and "one embodiment" mean "at least one example embodiment". The term "another embodiment" means "at least one additional embodiment". The terms "first," "second," and the like may refer to different or the same object.

As described above, in the conventional scheme for recommending a target object, a great gap exists between a recommended commodity and a commodity actually preferred by a user, and it is difficult to quickly and accurately match the user's preference with respect to the current commodity. Research shows that the operation characteristics such as the historical click rate of the user or the name of the commodity purchased by the user in the historical process cannot accurately reflect the characteristics of the commodity preferred by the user and cannot accurately reflect the actual characteristics of the commodity to be retrieved, so that the recommended commodity obtained based on the retrieval and the commodity actually preferred by the user have a large difference. In addition, the dimensionality of the user historical click features is high, and long calculation time is needed for retrieval and matching, so that recommended commodities cannot be obtained quickly.

To address, at least in part, one or more of the above problems and other potential problems, example embodiments of the present disclosure propose a scheme for recommending a target object. The scheme comprises the following steps: extracting image features of an image about a target object; obtaining comment information about the target object so as to generate a tag feature associated with the comment information based on the comment information; counting predetermined operations on the target object within a predetermined time interval to generate operation characteristics; fusing the image feature, the tag feature and the operation feature to generate a description feature about the target object, the dimension of the description feature being smaller than the sum of the dimensions of the image feature, the tag feature and the operation feature; determining the similarity between the current target object aimed by the user operation and the target object to be selected based on the description characteristics, and determining a display sequence for displaying the target object to be recommended to the user based on the similarity; and displaying the images of the target objects to be recommended based on the display sequence.

In the above scheme, the generated image feature, the tag feature and the operation feature about the target object are fused to generate a description feature about the target object, so that the description feature can accurately reflect the actual characteristics of the target object; in addition, by determining the similarity between the current target object and the target object to be selected, which is aimed at by the user operation, based on the description features, so as to determine the display sequence for displaying the target object to be recommended to the user based on the similarity, the present disclosure can enable candidate target objects to be more accurately matched with the current target object through similarity calculation based on the description features; in addition, by enabling the dimension of the description feature to be smaller than the sum of the dimensions of the image feature, the label feature and the operation feature, the retrieval and calculation speed of the description feature is facilitated to be improved, the target object to be recommended is determined quickly, and consumption or occupation of calculation and storage resources is reduced. Therefore, the present disclosure can quickly make the recommended target object accurately match the user's preference with respect to the current target object.

Fig. 1 shows a schematic diagram of a system 100 for implementing a method for recommending a target object according to an embodiment of the present disclosure. As shown in fig. 1, the system 100 includes: user terminal 110, computing device 130, server 160, and network 150. User terminal 110, computing device 130, server 160 may interact with data via network 150. The computing device 130 includes, for example: an image feature determination unit 132, a label feature generation unit 134, an operation feature generation unit 136, a description feature generation unit 138, a similarity calculation and ranking unit 140, and a display unit 142.

The calculation unit 130 is configured to display an image of the target object to be recommended based on the current target object targeted by the user operation. Specifically, the calculation unit 130 is configured to extract an image feature of an image regarding the target object; generating a tag feature associated with the comment information based on the comment information; generating an operating characteristic; fusing the image feature, the tag feature, and the operation feature to generate a descriptive feature about the target object; and determining the similarity between the current target object and the target object to be selected, which is aimed at by the user operation, based on the description characteristics, so as to determine the display sequence for displaying the target object to be recommended to the user and display the image of the target object to be recommended based on the similarity. The computing unit 130 may be a computing device, such as a server, having one or more processing units, including special purpose processing units, such as GPUs, FPGAs, and ASICs, and general purpose processing units, such as CPUs. In addition, one or more virtual machines may also be running on each computing device.

And an image-feature-related determination unit 132 for extracting an image feature of the image related to the target object.

A tag-related feature generating unit 134 for acquiring comment information on the target object so as to generate a tag feature associated with the comment information based on the comment information.

And an operation feature generation unit 136 for counting predetermined operations with respect to the target object within a predetermined time interval so as to generate an operation feature.

A description feature generation unit 138 for fusing the image feature, the tag feature and the operation feature to generate a description feature for the target object, and making a dimension of the description feature smaller than a sum of dimensions of the image feature, the tag feature and the operation feature.

And a similarity calculation and sorting unit 140, configured to determine, based on the description features, a similarity between the current target object and a target object to be selected, to which the user operates, and determine, based on the similarity, a display order in which the target object to be recommended is displayed to the user.

And a display unit 142 for displaying the image of the target object to be recommended based on the display order.

A method 200 for recommending a target object according to an embodiment of the present disclosure will be described below in conjunction with fig. 2. FIG. 2 shows a flow diagram of a method 200 for recommending a target object, according to an embodiment of the present disclosure. It should be understood that the method 200 may be performed, for example, at the electronic device 600 depicted in fig. 6. May also be executed at the computing device 130 depicted in fig. 1. It should be understood that method 200 may also include additional acts not shown and/or may omit acts shown, as the scope of the disclosure is not limited in this respect.

At step 202, the computing device 130 extracts image features about the image of the target object. The target object is, for example, commodity and content information.

In some embodiments, the computing device 130 extracts image features of each of all of the target objects based on the third neural network model. The third neural network model may be a CNN neural network structure. By adopting the CNN neural network structure, the image of the target object with large data volume can be effectively reduced into small data volume, and the storage and subsequent processing of the image characteristics of the target object are facilitated. In some embodiments, the third neural network model may also be other convolutional neural networks.

At step 204, the computing device 130 obtains review information regarding the target object to generate a tag feature associated with the review information based on the review information.

In some embodiments, computing device 130 may identify, based on the review information associated with the target object, an attribute about the review information and an emotional propensity for the attribute; determining comment information to be selected based on the determined emotional tendency about the attribute and the comment information; and then generating label characteristics based on the attributes of the comment information to be selected and the keyword information related to the attributes. The attribute for example comprises at least one of smell, taste, price, colour, size, function, quality, trim, electricity consumption situation, material, price. The method 400 for generating the tag feature will be described in detail with reference to fig. 4, and will not be described herein again.

At step 206, the computing device 130 counts the predetermined operations with respect to the target object over a predetermined time interval to generate operational characteristics. The predetermined operation on the target object includes at least one of: click operations for the target object, add shopping cart operations, place orders, set to focus, and so forth.

In some embodiments, the computing device 130 clusters the operational information for the target object in a predetermined time interval, for example, based on the target object, to generate operational characteristics. The operation characteristics include, for example, each operation type and the number of operations under each operation type.

At step 208, the computing device 130 fuses the image feature, the tag feature, and the operational feature to generate a descriptive feature about the target object, the descriptive feature having a dimension less than a sum of dimensions of the image feature, the tag feature, and the operational feature.

In some embodiments: the computing device 130 fusing the image features, the tag features, and the operational features to generate descriptive features about the target object includes: stitching the image features, the label features and the operation features to generate input features for inputting the first neural network model; and determining descriptive features for the target object via a first neural network model based on the input features, the first neural network model being trained via a plurality of samples.

In some embodiments, the manner in which the input features are generated includes, for example: the computing device 130 first normalizes the image features, the tag features, and the operation features, and then concatenates the normalized image features, the tag features, and the operation features to generate the input features. By normalizing the data, the optimization process of the optimal solution obviously becomes gentle, and the optimal solution is easier to be converged correctly.

In some embodiments, the first neural network model is constructed, for example, based on an auto-encoder. A self-coding network is an unsupervised learning algorithm, for example a 3-layer or greater than 3-layer neural network, which can give a better characterization than the original input data. For example, fig. 3 shows a schematic diagram of a first neural network model for generating descriptive features about a target object, in accordance with an embodiment of the present disclosure. As shown in fig. 3, the first neural network model is used to encode the input features (expression x) as the descriptive features (expression y) of the target object, and then decode the descriptive features (expression y) into expression (expression x') in order to train the network through a back propagation algorithm to make the output equal to the input. The processing of the first neural network model will be described below with reference to equations (1) to (3) and fig. 3.

y = f(x) = s(wx+b) （1）

x' = g(y) = s(w'y+b') （2）

L(x,x') = L(x,g(f(x))) （3）

In the above equations (1) to (3), wherein L represents a loss function. w' represents the first weight and w represents the second weight. x represents an input feature (e.g., input feature of input layer 310 in fig. 3). y represents a description feature (e.g., a feature of the middle layer 320 of fig. 3 via an encoding process). x' represents the decoded features (e.g., features of output layer 330 in fig. 3). f () represents an encoding processing function. g () represents a decode processing function. The loss function L may be a quadratic error (squared error) or a cross entropy error (cross entropy).

As can be seen from fig. 3, the dimension of the middle layer 320 is smaller than the dimension of the input feature x of the input layer 310. That is to say from the input feature x via a dimension reduction of the encoding process into the description feature y. Since the input feature x is formed by splicing the image feature, the tag feature and the operation feature, the input feature x has a higher latitude, and therefore, the original higher-dimension input feature x can be described by the description feature y with a smaller dimension through the conversion of the first neural network model 300 without losing rich information of the original input feature x, thereby facilitating the improvement of the retrieval and calculation speed of the description feature, further quickly determining a target object to be recommended, and reducing the consumption or occupation of calculation and storage resources.

At step 210, the computing device 130 determines similarity between the current target object and the target object to be selected, to which the user operates, based on the description features, so as to determine a display order for displaying the target object to be recommended to the user based on the similarity.

At step 212, the computing device 130 displays an image of the target object to be recommended based on the display order.

In the above scheme, by fusing the generated image feature, the tag feature and the operation feature about the target object, the description feature which accurately reflects the actual characteristics of the target object and has a smaller dimension can be generated, and based on the description feature, the similarity between the current target object and the target object to be selected, to which the user operates, is determined for determining the display order for displaying the target object to be recommended to the user based on the similarity, the present disclosure can quickly enable the recommended target object to accurately match the preference of the user about the current target object.

A method 400 for generating tag features according to an embodiment of the present disclosure will be described below in conjunction with fig. 4. Fig. 4 shows a flow diagram of a method 400 for generating tag characteristics, in accordance with an embodiment of the present disclosure. It should be understood that the method 400 may be performed, for example, at the electronic device 600 depicted in fig. 6. May also be executed at the computing device 130 depicted in fig. 1. It should be understood that method 400 may also include additional acts not shown and/or may omit acts shown, as the scope of the disclosure is not limited in this respect.

At step 402, the computing device 130 converts the comment information associated with the target object into a feature vector. The comment information is, for example, unstructured comment text, and therefore needs to be converted into a feature vector.

At step 404, the computing device 130 determines attribute probabilities for the review information based on the feature vectors, the attributes being associated with the type of the target object.

For example, the computing device 130 predicts the attribute probability about the comment information based on the feature vector converted from the comment information via a multi-sample trained attribute recognition model (which is, for example, built by the second neural network model prediction). For example, the comment information about a certain lipstick is "the lipstick is very fashionable in color, can brighten the face well, and is fresh in smell". The attributes of the comment information about the lipstick predicted by the second neural network model and ranked first two attribute probabilities are "color" and "smell". For another example, the comment information about a certain brand of apple is "the taste of his apple is sweet and delicious, and children like it, and the price is substantial". The top two attribute probabilities of the comment information about this brand of apple predicted by the second neural network model are, for example, "taste" and "price". The diverse examples used to train the second neural network model are, for example, a plurality of comment information via manual or automatic labeling of attributes.

At step 406, the computing device 130 determines attributes for the review information based on the determined attribute probabilities. For example, the computing device 130 determines attributes about the review information based on the attribute probability magnitudes. For example, the computing device 130 determines the attribute with the highest probability of the attribute as the attribute about the comment information. For example, an attribute of comment information about the aforementioned lipstick is determined as "color"; the attribute of the comment information about the aforementioned apple is determined as "taste".

At step 408, computing device 130 determines an emotional propensity probability for the attribute based on the review information for determining emotional propensity for the attribute.

For example, computing device 130 extracts features of the review information via a multi-sample trained emotional propensity recognition model (e.g., constructed based on a regression or classification model), predicting emotional propensity probabilities for the attributes. For example, the computing device 130 constructs a feature vector representation of a plurality of pieces of comment information, trains the regression or classification model based on the emotional tendency label information of the words in the emotional dictionary, obtains the trained regression or classification model, and predicts the new comment information to obtain emotional tendency label information of the new comment information. The analysis target of the emotion tendency recognition model is, for example, the probability of emotion tendency (for example, "positive direction" or "negative direction") with respect to the attribute in the given comment information. For example, the attribute of the comment information on lipstick mentioned above is "color", and the emotional tendency probability with respect to the attribute of "color" identified by the emotional tendency identification model is "forward" with a high emotional tendency probability. Computing device 130 determines that the emotional tendency with respect to attribute "color" is "forward" based on the "forward" direction for which the predicted probability of emotional tendency is higher.

The method for emotional tendency of the attribute further includes, for example: determining emotional tendencies for the attributes based on the constructed emotion dictionary (sentient lexicon). Based on the constructed emotion dictionary, emotion information can be given to the words. The affective information can be expressed, for example, as { positive, negative, neutral }. The emotion information identification method can also be determined by adopting a value-area-dominance (VAD) model or an Evaluation-probability-activity (EPA) model. Constructing the emotion dictionary can be via manual annotation or automated annotation.

At step 410, computing device 130 determines candidate review information based on the determined emotional tendencies and review information for the attributes.

In some embodiments, the method for determining the comment information to be selected includes, for example: if computing device 130 determines that the emotional tendency of the attribute meets a predetermined condition, computing device 130 determines whether the length of the comment information is greater than or equal to a predetermined length threshold; and if the computing device 130 determines that the length of the comment information is greater than or equal to a predetermined length threshold, determining that the comment information is the comment information to be selected. For example, computing device 130 first selects the emotional tendency of the attribute as "forward".

At step 412, the computing device 130 extracts keyword information associated with the emotional tendency of the attribute in the candidate review information to generate a tag feature associated with the target object based on the attribute and the keyword information.

A method 500 for determining a display order of target objects to be recommended according to an embodiment of the present disclosure will be described below with reference to fig. 5. Fig. 5 shows a flowchart of a method 500 for determining a display order of target objects to be recommended according to an embodiment of the present disclosure. It should be understood that the method 500 may be performed, for example, at the electronic device 600 depicted in fig. 6. May also be executed at the computing device 130 depicted in fig. 1. It should be understood that method 500 may also include additional acts not shown and/or may omit acts shown, as the scope of the disclosure is not limited in this respect.

At 502, the computing device 130 obtains a description of a current target object for which the user operates.

At 504, the computing device 130 performs a locality-sensitive hash calculation on the descriptive features of the current target object to generate a first hash value. By adopting the locality sensitive hash calculation, the description characteristics of the target object are remarkably reduced to a binary hash value with lower dimension. The advantage of using a locality sensitive hashing algorithm is also: the merging of the description characteristics of similar target objects can be completed without training, and the defects of the traditional algorithms such as clustering and the like that the data set is enlarged and the data set needs to be fitted again are not needed. In some embodiments, the computing device 130 may perform a hash calculation on the descriptive characteristics of the target object using a p-Stable (p-Stable) distributed LSH algorithm to generate a first hash value. The above-mentioned p stability profile can be described as: for one entityDistribution D over number set R, if P is present>=0, for any n real numbers v1, …, vn and n variables X1, …, Xn satisfying D distribution, random variables Σ iviXi (vector dot product, one number) and (Σ i | vi | y^p)^1/pX(P [ step of]Norm, a number) has the same distribution, where X is a random variable subject to the distribution of D, then D is called a p-stable distribution, for any p ∈ (0, 2)]There is a stable distribution: p =1 is the Cauchy distribution with a probability density function of c (x) =1/[ pi (1+ x2)](ii) a p =2 is a gaussian distribution and the probability density function is g (x) =1/(2 pi) 1/2 × e-x ^ 2/2.

And mapping the description characteristics of the target object into a hash value by using an LSH algorithm based on p stable distribution. The hash function is locally sensitive, so if the feature vector v1 describing the feature of the current target object and the feature vector v2 of the candidate target object are close to each other, the probability that the hash values mapped by the hash function will be the same and be hashed into the same hash bucket will be high.

The algorithm of the above locality sensitive hash calculation is described below with reference to equation (4).

hash_vector = A * v （4）

In the above equation (4), the hash _ vector represents a hash vector (or referred to as "hash value") calculated via the locality sensitive hash. A represents a hash function mapping matrix. v represents a feature vector of an image generated via features extracted by the neural network model. v is, for example, a feature vector describing features of the current target object for which the user operates, or a feature vector describing features of the target object stored in the database. And (4) judging the elements in the feature vector v through the hash function mapping matrix A shown in the formula (4), wherein if the elements are larger than 0, the corresponding features of the hash vector hash _ vector are 1, and otherwise, the corresponding features are 0.

An example of the hash function mapping matrix a is described below, for example, in connection with equation (5).

A=Shape：（N，D）（5）

In the above equation (5), N represents a mapping dimension of a hash function mapping function, i.e., a hash vector (or referred to as "hash value");) Dimension of hash vector. D represents the dimension of the feature vector v of the image. Shape: () For example representing a locality sensitive hash function based on a stable distribution of p. According to the p stable distribution, the Hash mapping distance A v1-A v2 and | v1-v2| of feature vectors v1 and v2 of the description features of two target objects^pThe distribution of X is the same. Therefore, the locally sensitive hash calculation of the p stable distribution can be used for effectively approximating the description characteristics of the target object.

At 506, the computing device 130 obtains the descriptive characteristics of the candidate target object having the same packet hash value as the first hash value, the packet hash value being generated via a locality sensitive hash calculation for the descriptive characteristics of the target object.

At 508, the computing device 130 respectively computes similarities of the descriptive features of the current target object and the descriptive features of the target objects to be selected, so as to determine an order of displaying the target objects to the user based on the similarity computation results.

In some embodiments, the method 500 further comprises: performing locality sensitive hash calculation on the description characteristics of each target object in all target objects to generate a packet hash value of each target object; multiple target objects having the same packet hash value are associated with the same hash bucket.

In the above scheme, the describing feature of the target object is subjected to hash mapping to a compact first hash value by a describing feature method of the target object based on locality sensitive hash calculation, so that resources consumed by calculation and storage can be greatly reduced. Therefore, the method and the device can effectively reduce the calculation and storage resources required by similarity calculation and matching, and improve and quickly enable the recommended target object to accurately match the current target object.

FIG. 6 schematically illustrates a block diagram of an electronic device (or computing device) 600 suitable for use to implement embodiments of the present disclosure. The apparatus 600 may be an apparatus for implementing the

methods

200, 400 to 500 shown in fig. 2, 4 to 5. As shown in fig. 6, device 600 includes a Central Processing Unit (CPU) 601 that may perform various appropriate actions and processes in accordance with computer program instructions stored in a Read Only Memory (ROM) 602 or loaded from a storage unit 608 into a Random Access Memory (RAM) 603. In the RAM603, various programs and data required for the operation of the device 600 can also be stored. The CPU601, ROM 602, and RAM603 are connected to each other via a bus 604. An input/output (I/O) interface 605 is also connected to bus 604.

A number of components in the device 600 are connected to the I/O interface 605, including: input unit 606, output unit 607, storage unit 608, processing unit 601 performs the various methods and processes described above, e.g., performs

methods

200, 400 to 500-e.g., in some embodiments,

methods

200, 400 to 500 may be implemented as a computer software program stored on a machine readable medium, e.g., storage unit 608. In some embodiments, part or all of the computer program may be loaded and/or installed onto the device 600 via the ROM 602 and/or the communication unit 609. When the computer program is loaded into RAM603 and executed by CPU601, one or more of the operations of

methods

200, 400 through 500 described above may be performed. Alternatively, in other embodiments, CPU601 may be configured by any other suitable means (e.g., by way of firmware) to perform one or more acts of methods 200, 400-500.

It should be further appreciated that the present disclosure may be embodied as methods, apparatus, systems, and/or computer program products. The computer program product may include a computer-readable storage medium having computer-readable program instructions embodied thereon for carrying out various aspects of the present disclosure.

The computer readable storage medium may be a tangible device that can hold and store the instructions for use by the instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic memory device, a magnetic memory device, an optical memory device, an electromagnetic memory device, a semiconductor memory device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a Static Random Access Memory (SRAM), a portable compact disc read-only memory (CD-ROM), a Digital Versatile Disc (DVD), a memory stick, a floppy disk, a mechanical coding device, such as punch cards or in-groove projection structures having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media as used herein is not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., optical pulses through a fiber optic cable), or electrical signals transmitted through electrical wires.

The computer-readable program instructions described herein may be downloaded from a computer-readable storage medium to a respective computing/processing device, or to an external computer or external storage device via a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. The network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in the respective computing/processing device.

The computer program instructions for carrying out operations of the present disclosure may be assembler instructions, Instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, the electronic circuitry that can execute the computer-readable program instructions implements aspects of the present disclosure by utilizing the state information of the computer-readable program instructions to personalize the electronic circuitry, such as a programmable logic circuit, a Field Programmable Gate Array (FPGA), or a Programmable Logic Array (PLA).

Various aspects of the present disclosure are described herein with reference to flowchart illustrations and/or step diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each step of the flowchart and/or step diagrams, and combinations of steps in the flowchart and/or step diagrams, can be implemented by computer-readable program instructions.

These computer-readable program instructions may be provided to a processor in a voice interaction device, a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processing unit of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or step diagram block or blocks. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer-readable medium storing the instructions comprises an article of manufacture including instructions which implement the function/act specified in the flowchart and/or step diagram step or steps.

The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or step diagram step or steps.

The flowcharts and step diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or step diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two successive method steps may in fact be executed substantially concurrently, or they may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each step of the step diagrams and/or flowchart illustration, and combinations of steps in the step diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

Having described embodiments of the present disclosure, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen in order to best explain the principles of the embodiments, the practical application, or improvements made to the technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

The above are merely alternative embodiments of the present disclosure and are not intended to limit the present disclosure, which may be modified and varied by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure should be included in the protection scope of the present disclosure.

Claims

1. A method for recommending a target object, comprising:

extracting image features of an image about a target object;

obtaining comment information about the target object so as to generate a tag feature associated with the comment information based on the comment information;

counting predetermined operations on the target object within a predetermined time interval so as to generate operation characteristics;

fusing the image feature, the tag feature, and the operation feature to generate a description feature for the target object, the dimension of the description feature being smaller than the sum of the dimensions of the image feature, the tag feature, and the operation feature;

determining similarity between the current target object aimed by the user operation and a target object to be selected based on the description characteristics, and determining a display sequence for displaying the target object to be recommended to the user based on the similarity; and

and displaying the images of the target objects to be recommended based on the display sequence.

2. The method of claim 1, wherein fusing the image features, the tag features, and the operational features to generate descriptive features about the target object comprises:

stitching the image features, the label features, and the operational features to generate input features for inputting into a first neural network model; and

determining the descriptive feature for the target object via the first neural network model based on the input features, the first neural network model being trained via a plurality of samples.

3. The method of claim 2, wherein the first neural network model is constructed based on a self-encoding network.

4. The method of claim 1, wherein generating a tag feature associated with the review information based on the review information comprises:

converting comment information associated with the target object into a feature vector;

determining attribute probabilities for the review information based on the feature vectors, the attributes being associated with the type of the target object;

determining attributes about the review information based on the determined attribute probabilities;

determining, based on the review information, an emotional tendency probability for the attribute for determining an emotional tendency for the attribute;

determining comment information to be selected based on the determined emotional tendency about the attribute and the comment information; and

extracting keyword information associated with the emotional tendency of the attribute in the comment information to be selected so as to generate the tag feature associated with the target object based on the attribute and the keyword information.

5. The method of claim 4, wherein the attributes include at least one of smell, taste, price, color, size, function, quality, interior, electricity consumption, material, price, the target object being a commodity.

6. The method of claim 4, wherein determining candidate review information based on the determined emotional tendencies for the attributes and the review information comprises:

in response to determining that the emotional tendency of the attribute meets a predetermined condition, determining whether the length of the comment information is greater than or equal to a predetermined length threshold; and

and in response to determining that the length of the comment information is greater than or equal to a preset length threshold, determining that the comment information is the comment information to be selected.

7. The method of claim 1, wherein determining, based on the descriptive features, a similarity to a current target object for which a user operation is directed and a target object to be selected for determining, based on the similarity, a display order in which to display target objects to be recommended to the user comprises:

obtaining the description characteristics of the current target object aimed at by the user operation;

performing locality sensitive hash calculation on the description features of the current target object to generate a first hash value;

obtaining the description characteristics of a target object to be selected, wherein the packet hash value is the same as the first hash value, and the packet hash value is generated by performing locality sensitive hash calculation on the description characteristics of the target object; and

and respectively calculating the similarity of the description features of the current target object and the description features of the target object to be selected so as to determine the sequence of displaying the target objects to the user based on the similarity calculation result.

8. The method of claim 7, further comprising:

performing locality sensitive hash calculation on the description characteristics of each target object in all target objects to generate a packet hash value of each target object; and

multiple target objects having the same packet hash value are associated with the same hash bucket.

9. The method of claim 1, wherein the predetermined operation with respect to the target object comprises at least one of: and clicking operation, shopping cart adding operation, ordering operation and setting as attention aiming at the target object.

10. A computing device, comprising:

at least one processing unit;

at least one memory coupled to the at least one processing unit and storing instructions for execution by the at least one processing unit, the instructions when executed by the at least one processing unit, cause the apparatus to perform the steps of the method of any of claims 1 to 9.

11. A computer-readable storage medium, having stored thereon a computer program which, when executed by a machine, implements the method of any of claims 1-9.