CN116029793A

CN116029793A - Commodity recommendation method, device, equipment and medium thereof

Info

Publication number: CN116029793A
Application number: CN202310082420.8A
Authority: CN
Inventors: 王耿鑫
Original assignee: Guangzhou Huaduo Network Technology Co Ltd
Current assignee: Guangzhou Huaduo Network Technology Co Ltd
Priority date: 2023-02-02
Filing date: 2023-02-02
Publication date: 2023-04-28

Abstract

The application relates to a commodity recommendation method, a device, equipment and a medium thereof in the technical field of electronic commerce, wherein the method comprises the following steps: acquiring dialogue text input by a user, and judging whether the dialogue text has a purchase intention or not; when the dialog text has purchase intention, a preset named entity model is adopted to identify commodity entity words and entity types thereof in the dialog text; matching expanded entity words synonymous with the commodity entity words from a preset synonymous entity word set according to the entity types of the commodity entity words; and splicing the commodity entity words and the expanded entity words to serve as search texts, and searching out corresponding recommended commodities according to the search texts. The method and the device effectively solve the dilemma that the social electronic commerce scene lacks a convenient commodity browsing page, can provide convenient commodity recommendation reply suggestions for merchants, improve shopping guide service efficiency of the merchants, and promote commodity transaction.

Description

Commodity recommendation method, device, equipment and medium thereof

Technical Field

The present disclosure relates to the field of electronic commerce technologies, and in particular, to a commodity recommendation method, and a corresponding apparatus, computer device, and computer readable storage medium thereof.

Background

In the scenes of social electronic commerce such as live shopping, text sales, cross-channel chat shopping and the like, customers need to communicate with merchants due to the lack of convenient commodity browsing pages, and in the communication process, the customers timely perceive the purchasing intention of the customers, determine corresponding commodities and feed the commodities back to the customers, so that it is important to guide the customers to promote the purchase.

In the conventional technology, the corresponding commodity searching content can be determined by using the dialogue content of the customer as input through the end-to-end deep learning model, however, in order to enable the deep learning model to have such capability, extremely huge dialogue content needs to be prepared, and corresponding commodity searching content is marked for each dialogue content, so that the deep learning model is supervised and trained to be in a convergence state. On the one hand, the training of the deep learning model to a convergence state requires a large amount of time, the finally obtained deep learning model is quite large and inconvenient to deploy the deep learning model to provide services online, and on the other hand, the marking of the dialogue content is usually realized manually, which means that huge manpower resources and time are required and the economic benefit is not met completely.

In view of the defects of the traditional technology, the applicant has long been engaged in research in the related field, and is in order to solve the problem in the field of electronic commerce, so a new way is developed.

Disclosure of Invention

It is a primary object of the present application to solve at least one of the above problems and provide a commodity recommendation method and corresponding apparatus, computer device, computer readable storage medium.

In order to meet the purposes of the application, the application adopts the following technical scheme:

a commodity recommendation method provided in accordance with one of the objects of the present application, comprising the steps of:

acquiring dialogue text input by a user, and judging whether the dialogue text has a purchase intention or not;

when the dialog text has purchase intention, a preset named entity model is adopted to identify commodity entity words and entity types thereof in the dialog text;

matching expanded entity words synonymous with the commodity entity words from a preset synonymous entity word set according to the entity types of the commodity entity words;

and splicing the commodity entity words and the expanded entity words to serve as search texts, and searching out corresponding recommended commodities according to the search texts.

In a further embodiment, before obtaining the dialog text input by the user, the method further includes the following steps:

acquiring commodity titles of a plurality of commodities, determining commodity entity words and entity types of the commodity entity words in each commodity title, and constructing commodity entity word sets corresponding to a plurality of entity types;

Acquiring a dialogue text input by a user history as a dialogue sample, labeling commodity entity words and entity types thereof related to purchase intention in the dialogue sample by adopting the commodity entity word set, and constructing a corresponding labeling sequence as a supervision label of the dialogue sample;

and training the named entity model to a convergence state by adopting the dialogue sample and the supervision label thereof, so that the named entity model is suitable for identifying commodity entity words and entity types thereof in dialogue texts.

acquiring a plurality of dialogue texts input by a user in a history way, and identifying commodity entity words and entity types of the commodity entity words in each dialogue text by adopting a preset named entity model;

splicing commodity entity words and entity types thereof as compound words to replace original commodity entity words in corresponding dialogue texts, and training a text coding model to a convergence state by adopting the replaced dialogue texts so as to ensure that the text coding model is suitable for determining word vectors corresponding to each word element in the dialogue texts;

determining word vectors corresponding to compound words formed by splicing each commodity entity word and entity type thereof by adopting the text coding model trained until convergence, calculating the similarity between each compound word, and aggregating compound words with the similarity exceeding a preset threshold value to obtain a plurality of synonymous compound word clusters;

And classifying all the synonymous compound word clusters according to the entity types of compound words in the compound word clusters to obtain synonymous commodity entity words corresponding to each identical entity type in the same genus to form a synonymous entity word set.

In a further embodiment, obtaining a dialogue text input by a user, and determining whether the dialogue text has a purchase intention includes:

and acquiring a dialogue text input by a user, extracting deep semantic information of the dialogue text by adopting a preset text classification model, acquiring corresponding semantic feature vectors for two classifications, and correspondingly determining whether the dialogue text has purchase intention or not according to classification results.

obtaining a single dialog sample and a supervision tag thereof from the prepared training set, the supervision tag characterizing whether the dialog sample has a purchase intention;

extracting deep semantic information of the dialogue text by adopting a text classification model, obtaining corresponding semantic feature vectors for two-classification, and obtaining corresponding classification results;

and determining a loss value of the classification result by adopting the supervision tag of the dialogue sample, updating the weight of the text classification model when the loss value does not reach a corresponding preset threshold value, and continuously calling other dialogue samples to perform iterative training until the text classification model converges.

In a further embodiment, the method includes the steps of, after the entity words of the commodity and the expanded entity words are spliced to serve as search texts and corresponding recommended commodities are searched out according to the search texts:

acquiring the click rate and conversion rate of each recommended commodity, matching the click rate and the conversion rate with corresponding weights, and calculating the recommendation score of each recommended commodity;

and screening the recommended commodity with the highest recommendation score as a high-quality commodity to carry out commodity recommendation.

acquiring sales quantity of the recommended commodities corresponding to each sales area, and determining the recommended commodity with the largest sales quantity corresponding to each sales area;

and acquiring the geographic position information of the user, and determining the recommended commodity with the largest sales volume in the sales area matched with the geographic position information to recommend the commodity.

On the other hand, the commodity recommendation device provided by adapting to one of the purposes of the application comprises an intention judging module, a model identifying module, a synonymous matching module and a commodity searching module, wherein the intention judging module is used for acquiring a dialogue text input by a user and judging whether the dialogue text has a purchase intention or not; the model identification module is used for identifying commodity entity words and entity types thereof in the dialogue text by adopting a preset named entity model when the dialogue text has purchase intention; the synonym matching module is used for matching the expanded entity words synonymous with the commodity entity words from a preset synonym entity word set according to the entity types of the commodity entity words; and the commodity searching module is used for splicing the commodity entity words and the expanded entity words to serve as search texts, and searching out corresponding recommended commodities according to the search texts.

In a further embodiment, before the intention judging module, the method further includes: the word set construction module is used for acquiring commodity titles of a plurality of commodities, determining commodity entity words and entity types thereof in each commodity title, and constructing commodity entity word sets corresponding to a plurality of entity types; the training preparation module is used for acquiring dialogue texts input by a user in history as dialogue samples, labeling commodity entity words and entity types thereof related to purchase intention in the dialogue samples by adopting the commodity entity word sets, and constructing corresponding labeling sequences as supervision labels of the dialogue samples; and the first model training module is used for training the named entity model to a convergence state by adopting the dialogue sample and the supervision label thereof, so that the named entity model is suitable for identifying commodity entity words and entity types thereof in the dialogue text.

In a further embodiment, before the intention judging module, the method further includes: the entity recognition module is used for acquiring a plurality of dialogue texts input by a user in a history way, and recognizing commodity entity words and entity types thereof in each dialogue text by adopting a preset named entity model; the second model training module is used for splicing commodity entity words and entity types thereof as compound words to replace original commodity entity words in corresponding dialogue texts, and training a text coding model to a convergence state by adopting the replaced dialogue texts so as to ensure that the text coding model is suitable for determining word vectors corresponding to each word element in the dialogue texts; the word cluster construction module is used for determining word vectors corresponding to compound words formed by splicing each commodity entity word and entity type thereof by adopting the text coding model trained until convergence, calculating the similarity between each compound word, and aggregating compound words with the similarity exceeding a preset threshold value to obtain a plurality of synonymous compound word clusters; and the word set forming module is used for classifying all the synonymous compound word clusters according to the entity types of compound words in the compound word clusters to obtain synonymous commodity entity words corresponding to the same entity types of the same genus to form the synonymous entity word set.

In a further embodiment, the intention judgment module includes: and the classification judging sub-module is used for acquiring the dialogue text input by the user, extracting the deep semantic information of the dialogue text by adopting a preset text classification model, acquiring corresponding semantic feature vectors for carrying out two classifications, and correspondingly determining whether the dialogue text has the purchase intention or not according to the classification result.

In a further embodiment, before the intention judging module, the method further includes: the sample acquisition module is used for acquiring a single dialogue sample and a supervision label thereof from the prepared training set, wherein the supervision label characterizes whether the dialogue sample has a purchase intention or not; the classification module is used for extracting deep semantic information of the dialogue text by adopting a text classification model, obtaining corresponding semantic feature vectors for carrying out two classification, and obtaining corresponding classification results; and the iteration training module is used for determining a loss value of the classification result by adopting a supervision tag of the dialogue sample, updating the weight of the text classification model when the loss value does not reach a corresponding preset threshold value, and continuously calling other dialogue samples to perform iteration training until the text classification model converges.

In a further embodiment, after the commodity searching module, the commodity searching module further includes: the scoring calculation module is used for obtaining the click rate and the conversion rate of each recommended commodity, matching the click rate and the conversion rate with corresponding weights, and calculating the recommendation score of each recommended commodity; and the commodity optimization module is used for screening the recommended commodity with the highest recommendation score as a high-quality commodity to carry out commodity recommendation.

In a further embodiment, after the commodity searching module, the commodity searching module further includes: the sales crown determining module is used for obtaining sales volumes of the recommended commodities corresponding to each sales area and determining recommended commodities with the largest sales volumes corresponding to each sales area; and the region recommending module is used for acquiring the geographic position information of the user and determining recommended commodities with the largest sales volume in the sales region matched with the geographic position information to recommend the commodities.

In yet another aspect, a computer device is provided, adapted for one of the objects of the present application, comprising a central processor and a memory, the central processor being adapted to invoke the steps of running a computer program stored in the memory to perform the merchandise recommendation method described herein.

In yet another aspect, a computer readable storage medium adapted to another object of the present application is provided, in which a computer program implemented according to the commodity recommendation method is stored in the form of computer readable instructions, which computer program, when being invoked by a computer to run, performs the steps comprised by the method.

The technical solution of the present application has various advantages, including but not limited to the following aspects:

according to the method and the device, whether the dialog text input by the user has the purchase intention is judged, when the dialog text has the purchase intention, the commodity entity words and entity types thereof in the dialog text are identified by adopting the named entity model, further, the expanded entity words synonymous with the commodity entity words are matched from the synonymous entity word set according to the entity types of the commodity entity words, and the commodity entity words and the expanded entity words are spliced to serve as search texts to search out corresponding recommended commodities. The method and the system effectively solve the dilemma that the social electronic commerce scene lacks a convenient commodity browsing page, can provide convenient commodity recommendation reply suggestions for merchants, improve shopping guide service efficiency of the merchants, promote commodity transaction, and enable novice merchants to be capable of providing good shopping guide experience for customer users.

Drawings

The foregoing and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:

FIG. 1 is a flow chart of an exemplary embodiment of a merchandise recommendation method of the present application;

FIG. 2 is a flow chart of constructing training samples and supervising labels to train a named entity model to a converging state in an embodiment of the present application;

FIG. 3 is a flow diagram of constructing a set of synonyms entity in an embodiment of the present application;

FIG. 4 is a schematic diagram of a training process of a text classification model in an embodiment of the present application;

FIG. 5 is a schematic flow chart of preferred quality goods among recommended goods in the embodiment of the present application;

FIG. 6 is a flowchart of determining a recommended commodity with the largest sales volume in a sales area matching with geographic location information of a user according to an embodiment of the present application;

FIG. 7 is a schematic block diagram of a merchandise recommendation apparatus of the present application;

fig. 8 is a schematic structural diagram of a computer device used in the present application.

Detailed Description

Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary only for the purpose of illustrating the present application and are not to be construed as limiting the present application.

As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless expressly stated otherwise, as understood by those skilled in the art. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or wirelessly coupled. The term "and/or" as used herein includes all or any element and all combination of one or more of the associated listed items.

It will be understood by those skilled in the art that all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs unless defined otherwise. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

As used herein, "client," "terminal device," and "terminal device" are understood by those skilled in the art to include both devices that include only wireless signal receivers without transmitting capabilities and devices that include receiving and transmitting hardware capable of two-way communication over a two-way communication link. Such a device may include: a cellular or other communication device such as a personal computer, tablet, or the like, having a single-line display or a multi-line display or a cellular or other communication device without a multi-line display; a PCS (Persona l Commun i cat ions Service, personal communication system) that may combine voice, data processing, facsimile and/or data communication capabilities; a PDA (Persona l Digita l Ass i stant ) that can include a radio frequency receiver, pager, internet/intranet access, web browser, notepad, calendar and/or GPS (Gl oba l Pos it ion i ng System ) receiver; a conventional laptop and/or palmtop computer or other appliance that has and/or includes a radio frequency receiver. As used herein, "client," "terminal device" may be portable, transportable, installed in a vehicle (aeronautical, maritime, and/or land-based), or adapted and/or configured to operate locally and/or in a distributed fashion, at any other location(s) on earth and/or in space. As used herein, a "client," "terminal device," or "terminal device" may also be a communication terminal, a network access terminal, or a music/video playing terminal, for example, a PDA, an md (Mob I l eI nternet Dev I ce ), and/or a mobile phone with music/video playing function, or may also be a smart tv, a set top box, or other devices.

The hardware referred to by the names "server", "client", "service node" and the like in the present application is essentially an electronic device having the performance of a personal computer, and is a hardware device having necessary components disclosed by von neumann's principle, such as a central processing unit (including an arithmetic unit and a controller), a memory, an input device, and an output device, and a computer program is stored in the memory, and the central processing unit calls the program stored in the external memory to run in the memory, executes instructions in the program, and interacts with the input/output device, thereby completing a specific function.

It should be noted that the concept of "server" as referred to in this application is equally applicable to the case of a server farm. The servers should be logically partitioned, physically separate from each other but interface-callable, or integrated into a physical computer or group of computers, according to network deployment principles understood by those skilled in the art. Those skilled in the art will appreciate this variation and should not be construed as limiting the implementation of the network deployment approach of the present application.

One or several technical features of the present application, unless specified in the plain text, may be deployed either on a server to implement access by remotely invoking an online service interface provided by the acquisition server by a client, or directly deployed and run on the client to implement access.

The neural network model cited or possibly cited in the application can be deployed on a remote server and used for implementing remote call on a client, or can be deployed on a client with sufficient equipment capability for direct call unless specified in a clear text, and in some embodiments, when the neural network model runs on the client, the corresponding intelligence can be obtained through migration learning so as to reduce the requirement on the running resources of the hardware of the client and avoid excessively occupying the running resources of the hardware of the client.

The various data referred to in the present application, unless specified in the plain text, may be stored either remotely in a server or in a local terminal device, as long as it is suitable for being invoked by the technical solution of the present application.

Those skilled in the art will appreciate that: although the various methods of the present application are described based on the same concepts so as to be common to each other, the methods may be performed independently, unless otherwise indicated. Similarly, for each of the embodiments disclosed herein, the concepts presented are based on the same inventive concept, and thus, the concepts presented for the same description, and concepts that are merely convenient and appropriately altered although they are different, should be equally understood.

The various embodiments to be disclosed herein, unless the plain text indicates a mutually exclusive relationship with each other, the technical features related to the various embodiments may be cross-combined to flexibly construct a new embodiment, so long as such combination does not depart from the inventive spirit of the present application and can satisfy the needs in the art or solve the deficiencies in the prior art. This variant will be known to the person skilled in the art.

A commodity recommendation method of the present application may be programmed as a computer program product that is deployed to run in a client or server, for example, in an exemplary application scenario of the present application, may be deployed in a server of an e-commerce platform, whereby the method may be performed by accessing an interface that is open after the computer program product is run, and performing man-machine interaction with a process of the computer program product through a graphical user interface.

Referring to fig. 1, in an exemplary embodiment, the commodity recommendation method of the present application includes the following steps:

step S1100, acquiring dialogue text input by a user, and judging whether the dialogue text has a purchase intention or not;

in the scenes of social electronic commerce such as live shopping, text-attached sales, cross-channel chat shopping and the like, due to the lack of a convenient commodity browsing page, a corresponding chat portal is generally provided, so that customers can conveniently communicate with merchants through the chat portal, the users can communicate with the merchants in a dialogue mode through the corresponding chat window, in order to timely perceive the purchase intention of the users, so that the corresponding commodity is determined to be used as a recommended commodity of the merchant for replying the users, shopping guide of the merchants is assisted, and in the communication process, dialogue texts input by the users are acquired. The communication between the user and the merchant may be that the user consults the merchant about the physical distribution condition, the appearance, the use experience and the like of the commodity purchased by the user, so that the corresponding user's dialogue text does not have a purchase intention, and the user consults the merchant about the commodity characteristic, the use experience, the appearance and the like of the commodity which the user wants to purchase, so that the corresponding user's dialogue text has a purchase intention. For the dialogue text with no purchase intention, the merchant replies to the user according to the dialogue text without taking any further operation, however, for the dialogue text with purchase intention, the merchant needs to further determine the corresponding commodity, so as to judge whether the obtained dialogue text has purchase intention or not, and accordingly determine whether to perform the next operation.

In one embodiment, a preset text classification model may be used to determine whether the obtained dialog text has a purchase intention, specifically, the dialog text is used as an input of the text classification model, deep semantic information of the dialog text is extracted, corresponding semantic feature vectors are obtained and mapped to a classification space, the classification space includes a first classification space representing that the dialog text has the purchase intention and a second classification space not having the purchase intention, probabilities corresponding to the first classification space and the second classification space are determined, and a representation corresponding to the classification space with the highest probability is determined as a classification result, thereby, whether the dialog text has the purchase intention can be determined accordingly.

The preset text classification model is trained to a convergence state in advance, and the capability of determining whether the dialog text has the purchase intention is obtained. The text classification model may be constructed by a deep learning model based on deep semantic learning in the NLP (Natura l Language Process i ng) field adapted to extract text semantic features followed by a classifier adapted for classification tasks, e.g. LSTM, bi LSTM, or using an open source framework Sentence Transformers providing a number of pre-trained to converging convertor models, such as: bert, roBERTa, XLM-RoBERTa, MPNet, the specific choice can be flexibly realized by one skilled in the art.

Step 1200, when the dialog text has a purchase intention, identifying commodity entity words and entity types thereof in the dialog text by adopting a preset named entity model;

the named entity model is suitable for a named entity recognition task, the specific model selection can be RoBERTa+CRF, bi LSTM+CRF, IDCNN+CRF, bert+BiLSTM+CRF, FLAT and the like, one type selection can be selected according to the needs of a person skilled in the art to realize, and the named entity model can acquire the capability of recognizing commodity entity words and entity types in the dialogue text after training until convergence.

In one embodiment, a Bert+BiLSTM+CRF model is adopted as a named entity model, after the model is trained in advance until convergence, a dialogue text with purchase intention is input as the named entity model, a Bert is used as an embedd i ng layer to extract deep semantic information of the dialogue text, a corresponding text feature sequence is output and input to the BiLSTM layer, the text feature sequence comprises a feature vector corresponding to each single word or word in the dialogue text in a vectorization manner, the BiLSTM layer outputs a score corresponding to each single word or word in the dialogue text and belonging to each category as the input of the CRF layer, and the CRF layer outputs a category sequence, wherein the category sequence comprises the category corresponding to each single word or word in the dialogue text, and commodity entity words and entity types thereof in the dialogue text can be correspondingly determined according to the category sequence. The category is obtained by adopting one of BIO, BIOES, BMES labeling methods based on the entity type. The entity type is used for classifying various types of commodity entity words, and can be specifically set by a person skilled in the art according to requirements, and examples include category, brand, applicable crowd, production place, material, specification/model, style/color, other attributes and the like.

Step S1300, matching the expanded entity words synonymous with the commodity entity words from a preset synonymous entity word set according to the entity types of the commodity entity words;

since the user may only memorize part of the information of the commodity or the user only has rough impression about the commodity, and further when describing the commodity in a dialogue with a merchant, the synonym is expressed by using synonyms, wherein the synonym refers to the synonym expression corresponding to the description information of the commodity by the user, however, all refer to the same commodity, for example, the user wants the commodity to be a millet mobile phone, and when the user describes the millet mobile phone in the dialogue with the merchant, the synonym expression such as M I or X i aoM i or millet may be used. It can be understood that the corresponding commodity is recalled by the commodity entity words identified by the dialogue text of the user, and the situation that the recall is too few or even impossible may occur, so that the synonyms of the commodity entity words need to be expanded, and further, more corresponding commodities are reasonably and effectively recalled according to the commodity entity words and the expanded synonyms thereof. For this purpose, a synonym entity set including a plurality of synonym commodity entity subsets corresponding to each entity type is prepared, the entity type of the commodity entity is matched with the entity type in the synonym entity set, a plurality of synonym commodity entity subsets of the entity type matched in the synonym entity set are obtained, commodity entity words except the commodity entity words are obtained from the synonym commodity entity word subset which is determined to exist in the commodity entity words, the commodity entity words are synonymous with the commodity entity words, and the commodity entity words are used as extension entity words of the commodity entity words.

And step 1400, splicing the commodity entity words and the expanded entity words to serve as search texts, and searching out corresponding recommended commodities according to the search texts.

And splicing the commodity entity words and the extended entity words by adopting space symbols, constructing a search text, searching commodity information of each commodity in a commodity database according to the search text, and searching out the commodity matched with the search text as a recommended commodity. The commodity database is created and maintained by a merchant of an online store, and the commodity information comprises but is not limited to text information and picture information of the commodity, wherein the text information can comprise a commodity title of the commodity, a commodity detail text, a commodity class, a commodity label and other texts describing the commodity. The picture information can comprise pictures which are uploaded for displaying the commodity when the commodity is put on shelf by a merchant of an online store, and the commodity can be displayed from the whole and/or different sides, including commodity main pictures, commodity detail pictures and the like. And when the text information and/or the picture information in the commodity information of the commodity in the commodity database is matched with the search text, determining that the commodity is matched with the search text.

On the one hand, for determining whether the search text is matched with the picture information of the commodity in the commodity database, a preset image-text matching model can be adopted to take one picture in the picture information of the commodity in the search text and the commodity database as input, deep semantic features corresponding to the search text and the picture are extracted, corresponding two semantic feature vectors are mapped to the same semantic space, the distance between the two semantic feature vectors is calculated to serve as similarity between the search text and the picture of the commodity in the commodity database, whether the search text is matched with the picture of the commodity in the commodity database is determined according to whether the similarity exceeds a preset threshold value or not correspondingly, and the similarity calculation can be implemented by any one of large-scale vector search engines such as Fai ss, E l ast i cSearch, mi l vus and the like, or can be calculated by adopting any one existing algorithm such as cosine similarity, inner product, manhattan distance and the like. The preset threshold value can be set by a person skilled in the art as required, the preset image-text matching model is trained to a convergence state in advance, the recommended model is a CLI P model, and the training process of the CLI P model is not described in detail because the training process of the CL I P model is known to the person skilled in the art.

On the other hand, aiming at determining whether the search text is matched with text information of the commodity in the commodity database, the text information of the corresponding commodity is searched out by fuzzy search or precision according to the search text, and the text information of the commodity is matched with the search text.

Acquiring commodity links of the recommended commodities to construct corresponding commodity cards, enabling users to reach display pages of the corresponding recommended commodities conveniently through touch control of the commodity cards, taking the commodity cards as intelligent reply suggestions of merchant reply users, optionally pushing the commodity cards to the users from the intelligent reply suggestions when the merchants reply, wherein the specific display forms of the pushed commodity cards can be flexibly changed by one skilled in the art,

as can be appreciated from the exemplary embodiments of the present application, the technical solution of the present application has various advantages, including but not limited to the following aspects:

Referring to fig. 2, in a further embodiment, before step S1100, the method further includes the following steps:

step S1000, acquiring commodity titles of a plurality of commodities, determining commodity entity words and entity types of the commodity entity words in each commodity title, and constructing commodity entity word sets corresponding to a plurality of entity types;

the commodity entity words in each commodity title can be manually determined by acquiring a data set of commodity titles of open-source electronic commerce commodities or acquiring commodity titles of commodities from a plurality of commodity databases, and the entity types are marked and used for classifying various types of commodity entity words, and can be specifically set by a person skilled in the art according to requirements, and examples of the entity types include categories, brands, applicable crowds, places of production, materials, specifications/models, styles/colors, other attributes and the like. And according to the entity types corresponding to each commodity entity word, collecting commodity entity words belonging to the same entity type, thereby constructing a commodity entity word set corresponding to a plurality of entity types.

Step S1001, a dialogue text input by a user history is obtained as a dialogue sample, commodity entity words and entity types thereof related to purchase intention in the dialogue sample are marked by adopting the commodity entity word set, and a corresponding marking sequence is constructed as a supervision tag of the dialogue sample;

It can be understood that the commodity title is significantly different from the dialogue text expressed by the spoken language of the user, that is, the commodity title cannot be used as a substitute for the dialogue text, so that the commodity title is not suitable for directly using the commodity title as a training sample, and instead, a certain scale of dialogue text input by the user in history is obtained.

Further, the commodity entity words and the entity types thereof in the dialogue sample need to be marked, the dialogue sample is subjected to word segmentation processing to obtain a corresponding word segmentation sequence, for the dialogue sample to be Chinese, algorithms such as j i eba, stanford, han l p, KCWS word segmentation device, THULAC, N-Gram, deep learning and the like can be adopted, and for the dialogue sample to be natural language separated by space between words such as English, french, german and the like as separators, the commodity titles can be segmented simply through a space segmentation mode, or algorithms such as Keras, spacy, gensim, NNTK, N-Gram, deep learning and the like can be adopted. And searching whether the commodity entity word sets of the plurality of entity types have commodity entity words which are the same as the word elements in the word segmentation sequence, determining corresponding commodity entity words in the word segmentation sequence, and determining the entity type of the commodity entity words in the word segmentation sequence according to the entity type corresponding to the commodity entity word set of the commodity entity words. It can be appreciated that the foregoing automatic labeling of the dialogue sample by using the set of commodity entity words may have the condition of missing label of the commodity entity words, and in addition, the dialogue sample may be doped with commodity entity words irrelevant to the purchase intention of the user, so that the labeled dialogue sample can be further supplemented and labeled manually for correction, for example, the dialogue sample is "is a piece of black lamp core velvet trousers washable by a washing machine? Is there a fade? The commodity entity words which can be automatically marked are black, corduroy, trousers and washing machine, the corresponding entity types are color, material, class and class, however, the washing machine is a commodity entity word which is irrelevant to the purchase intention of a user, and the marking of the commodity entity word can be corrected to be ignored. Based on the commodity entity words marked by the dialogue sample and the entity types thereof, one of BIO, BIOES, BMES marking methods is adopted, a corresponding marking sequence is marked as a supervision label of the dialogue sample, and one marking method can be selected by a person skilled in the art as required to realize marking.

Step S1002, training the named entity model to a convergence state by using the dialogue sample and the supervision label thereof, so that the named entity model is suitable for identifying commodity entity words and entity types thereof in the dialogue text.

In one embodiment, performance and execution efficiency of a named entity model are considered, a Roberta+CRF model is recommended to be used as the named entity model, the dialogue sample is input to the Roberta+CRF model, deep semantic features corresponding to each single word or word in the dialogue sample are extracted by a Roberta layer, corresponding feature vectors are obtained, each feature vector is mapped to a preset multi-classification space, a score belonging to each category corresponding to each single word or word in the dialogue sample is output as an input of the CRF layer, a prediction category sequence is output by the CRF layer, the prediction category sequence comprises categories corresponding to each single word or word in the dialogue text, and the categories are obtained by adopting one of BIO, BIOES, BMES labeling methods based on entity types. Further, a preset cross entropy loss function is called, the cross entropy loss function can be flexibly set by a person skilled in the art according to priori knowledge or experimental experience, a cross entropy loss value corresponding to the prediction category sequence is calculated according to a supervision label of the dialogue sample, namely a labeling sequence, and when the cross entropy loss value reaches a corresponding preset threshold value, the named entity model is trained to a convergence state, so that model training can be terminated; and when the cross entropy loss value does not reach the corresponding preset threshold value, indicating that the model is not converged, then carrying out gradient update on the model according to the cross entropy loss value, generally correcting the weight parameters of each link of the model through back propagation to enable the model to further approach convergence, and then continuously calling other dialogue samples to carry out iterative training on the named entity model until the named entity model is trained to a convergence state.

In this embodiment, by constructing a training sample of the named entity model and labeling the supervision label thereof, the named entity model is trained to a convergence state, and the accuracy of named entity recognition can be ensured and unnecessary time consumed by invalid recognition can be avoided because commodity entity words and entity types thereof related to purchase intention in the dialogue sample are labeled pertinently.

Referring to fig. 3, in a further embodiment, before step S1100, the method further includes the following steps:

step S1010, acquiring a plurality of dialogue texts input by a user history, and identifying commodity entity words and entity types thereof in each dialogue text by adopting a preset named entity model;

the method comprises the steps of obtaining dialogue texts input by a user history of a certain scale, taking a single dialogue text as input by a named entity model which is trained to be converged in advance, such as a RoBERTa+CRF model, extracting deep semantic features corresponding to each single word or word in the dialogue text by a RoBERTa layer, obtaining corresponding feature vectors, mapping each feature vector to a preset multi-classification space, outputting scores belonging to each category corresponding to each single word or word in the dialogue text as input of a CRF layer, outputting a category sequence by the CRF layer, and determining commodity entity words and entity types thereof in the dialogue text according to the category sequence.

Step S1020, splicing commodity entity words and entity types thereof as compound words to replace original commodity entity words in corresponding dialogue texts, and training a text coding model to a convergence state by adopting the replaced dialogue texts so as to ensure that the dialogue texts are suitable for determining word vectors corresponding to each word element in the dialogue texts;

it will be appreciated that in the e-commerce field, the same commodity entity word may belong to a plurality of different entity types, i.e. the word ambiguous, for example: the commodity entity words "millet" can belong to both brand entity types and category entity types, in order to define that the commodity entity words refer to a semantic meaning, the commodity entity words and the entity types thereof are spliced to form a composite word, the composite word can eliminate the divergence of the semantic meaning, and for each dialogue text, the composite word can replace the original commodity entity words in the corresponding dialogue text, the replaced dialogue text is used as a training sample to train a text coding model, the text coding model is a deep learning model based on deep semantic learning, which is suitable for extracting text semantic features in the NLP (Natura l Language Process i ng) field, the deep learning model is used for word2vec, LSTM, biLSTM or an open source framework Sentence Transformers is used, and a large number of pre-trained to converged converter models are provided, such as: bert, roBERTa, XLM-RoBERTa, MPNet, the specific choice can be flexibly realized by one skilled in the art. Since the training process of the text coding model is known to those skilled in the art, the training process is not described in detail.

Step 1030, determining word vectors corresponding to the compound words formed by splicing each commodity entity word and entity type thereof by using the text coding model trained until convergence, calculating the similarity between each compound word, and aggregating compound words with the similarity exceeding a preset threshold value to obtain a plurality of synonymous compound word clusters;

and extracting deep semantic features corresponding to the compound words formed by splicing each commodity entity word and entity types thereof by adopting the text coding model trained until convergence, coding corresponding word vectors, calculating vector distances between the word vectors corresponding to each compound word, and representing semantic similarity between the compound words by using the vector distances, wherein the calculation of the vector distances can be calculated by adopting any one existing algorithm such as cosine similarity, inner product, manhattan distance, euclidean distance and the like. The compound words with similarity exceeding the preset threshold are aggregated to obtain a corresponding plurality of synonymous compound word clusters, wherein the synonymous compound word clusters are exemplified by (millet, brand), (Xi aoMi, brand), (M I, brand) which is one synonymous word cluster, and (millet, category), (maize, category), (millet, category), (mi l et, category) which is another synonymous word cluster.

And step S1040, classifying all the synonymous compound word clusters according to the entity types of compound words in the compound word clusters to obtain synonymous commodity entity words corresponding to each same entity type in the same genus to form a synonymous entity word set.

It may be understood that, the commodity entity words corresponding to each compound word in the synonymous compound word cluster are synonymous, however, in order to facilitate determination of synonymous commodity entity words of the same entity type, the synonymous compound word clusters are categorized according to the entity types of the compound words, and for each entity type, the synonymous commodity entity words belonging to the same entity type in each synonymous compound word cluster are correspondingly collected to form a synonymous entity word set, where an exemplary example of the synonymous entity word set is as follows:

brand: { [ millet, X i aomi, M I ], [ apple, app l e ], … },

……

class: { [ millet, mi l et ], [ waistband, belt, gi rd l e, be t, sash ], … }

In the embodiment, the construction of the synonymous entity word set is disclosed, in the process, the commodity entity words are subjected to semantic disambiguation to refer to single semantics, and the same commodity entity words with different semantics are effectively distinguished, so that the accuracy and the reliability of the finally constructed synonymous entity word set are ensured.

Referring to fig. 4, in a further embodiment, before step S1100, the method further includes the following steps:

step S2000, acquiring a single dialogue sample and a supervision tag thereof from a prepared training set, wherein the supervision tag characterizes whether the dialogue sample has a purchase intention or not;

the method comprises the steps of obtaining a conversation text input by a user in a history of a certain scale in advance as a conversation sample, labeling a supervision label of the conversation sample, if the conversation sample content mentions commodity information of commodities and indicates that the commodities are wanted to be further known or potential or explicit shopping demands exist, labeling the supervision label of the conversation sample as a supervision label representing that buying intention exists in the conversation sample, and if the conversation sample content does not mention commodity information of the commodities or mentions commodity information of the commodities but does not need to be further known or the potential or explicit shopping demands exist, labeling the supervision label of the conversation sample as representing that buying intention does not exist in the conversation sample. Each dialogue sample is associated with its supervision tag to construct a training set.

Step S2100, extracting deep semantic information of the dialogue text by adopting a text classification model, obtaining corresponding semantic feature vectors for two-class classification, and obtaining corresponding classification results;

And taking the dialogue sample as input of a text classification model, extracting deep semantic information of the dialogue sample, obtaining corresponding semantic feature vectors, mapping the corresponding semantic feature vectors to a classification space, wherein the classification space comprises a first classification space representing that purchase intention exists and a second classification space representing that the purchase intention does not exist, determining the probability of mapping to the first classification space and the second classification space, and determining the corresponding characterization of the classification space with the maximum probability as a classification result.

The text classification model may be constructed by a deep learning model based on deep semantic learning in the NLP (Natura l Language Process i ng) field adapted to extract text semantic features followed by a classifier adapted to the classification task, e.g. LSTM, biLSTM, or using an open source framework Sentence Transformers that provides a number of pre-trained to converged converver models, such as: bert, roBERTa, XLM-RoBERTa, MPNet, the specific choice can be flexibly realized by one skilled in the art.

Step S2200, adopting a supervision tag of the dialogue sample to determine a loss value of the classification result, when the loss value does not reach a corresponding preset threshold value, updating the weight of the text classification model, and continuously calling other dialogue samples to perform iterative training until the text classification model converges.

Invoking a preset cross entropy loss function, wherein the cross entropy loss function can be flexibly set by a person skilled in the art according to priori knowledge or experimental experience, and according to a supervision tag of the dialogue sample, calculating a cross entropy loss value corresponding to the classification result, and when the cross entropy loss value reaches a corresponding preset threshold value, indicating that the text classification model is trained to a convergence state, so that model training can be terminated; and when the cross entropy loss value does not reach the corresponding preset threshold value, indicating that the model is not converged, carrying out gradient update on the model according to the cross entropy loss value, correcting the weight parameters of each link of the model through back propagation to enable the model to further approach convergence, and then continuously calling other dialogue samples to carry out iterative training on the text classification model until the text classification model is trained to a convergence state.

In this embodiment, training the text classification model to a convergence state is disclosed, and the ability to determine whether the dialog text has a purchase intention is learned. The dialogue text is input into the text classification model, so that whether the dialogue text has the purchasing intention or not can be quickly and accurately determined.

Referring to fig. 5, in a further embodiment, step S1400, the step of splicing the commodity entity word and the extended entity word as search text, and after searching out the corresponding recommended commodity according to the search text, further includes the following steps:

Step 1410, obtaining the click rate and conversion rate of each recommended commodity, matching the click rate and conversion rate with corresponding weights, and calculating the recommendation score of each recommended commodity;

the click rate is the ratio of the number of times the recommended commodity is clicked after exposure divided by the number of exposure, and the conversion rate is the ratio of the number of times the recommended commodity is purchased and/or added to a shopping cart divided by the number of times the recommended commodity is clicked to reach the display page of the commodity. Accordingly, the clicking rate and the conversion rate of each recommended commodity are obtained, the clicking rate and the conversion rate corresponding to each recommended commodity are multiplied by corresponding weights respectively and then added, corresponding recommendation scores are calculated, the weights corresponding to the average value of the clicking rate and the average value of the conversion rate respectively are added to be 1, the method can be flexibly set by a person skilled in the art, the weight corresponding to the recommended clicking rate is 0.4, and the weight corresponding to the conversion rate is 0.6.

And S1420, screening out the recommended commodity with the highest recommendation score as a high-quality commodity to recommend the commodity.

The recommended commodity with the highest recommendation score is the most popular among a plurality of recommended commodities, is most required by sales market, can be regarded as a good commodity, and is recommended to a user to be expected to achieve commodity transaction. Implementation of a specific merchandise recommendation may be revealed with reference to the corresponding portion of the intelligent reply recommendation of the construction merchant in step S1400.

In this embodiment, corresponding recommendation scores are obtained based on click rates and conversion rates of the recommended commodities, so that sales results of the corresponding recommended commodities are scientifically and effectively quantified and reflected, and accordingly, high-quality commodities with optimal sales results are screened and recommended to users, and commodity transactions are hopefully promoted.

Referring to fig. 6, in a further embodiment, step S1400, the step of splicing the commodity entity word and the extended entity word as search text, and after searching out the corresponding recommended commodity according to the search text, further includes the following steps:

step S1401, acquiring sales amounts of the recommended commodities corresponding to each sales area, and determining a recommended commodity with the largest sales amount corresponding to each sales area;

it will be appreciated that there may be a regional difference in sales of the products, and the sales volumes of the products sold in different sales areas may be different, so that in order to screen out recommended products most suitable for sales in the corresponding sales areas, the recommended products with the largest sales volumes corresponding to each sales area are counted, and the sales areas may be defined by those skilled in the art as required.

Step S1402, obtaining the geographical location information of the user, and determining the recommended commodity with the largest sales volume in the sales area matched with the geographical location information to perform commodity recommendation.

After the authorization of the user is obtained, the geographical position information of the terminal equipment operated by the user is obtained, and further, the recommended commodity with the largest sales volume in the sales area affiliated to the geographical position information is determined and recommended to the user to hopefully achieve commodity transaction. Implementation of a specific merchandise recommendation may be revealed with reference to the corresponding portion of the intelligent reply recommendation of the construction merchant in step S1400.

In this embodiment, by counting the recommended merchandise with the largest sales volume corresponding to each sales area, the recommended merchandise with the largest sales volume in the sales area matched with the geographical location information of the user is determined to be recommended to the user, and the recommended merchandise is expected to be promoted for commodity transaction.

Referring to fig. 7, a commodity recommendation device provided to adapt to one of the purposes of the present application is a functional implementation of the commodity recommendation method of the present application, where the device includes an intention judging module 1100, a model identifying module 1200, a synonym matching module 1300, and a commodity searching module 1400, where the intention judging module 1100 is configured to obtain a dialogue text input by a user, and determine whether the dialogue text has a purchase intention; the model identifying module 1200 is configured to identify, when the dialog text has a purchase intention, a commodity entity word and an entity type thereof in the dialog text by using a preset named entity model; the synonym matching module 1300 is configured to match, according to the entity type of the commodity entity word, an extended entity word synonymous with the commodity entity word from a preset synonym entity word set; the commodity searching module 1400 is configured to splice the commodity entity word and the expanded entity word as a search text, and search out a corresponding recommended commodity according to the search text.

In a further embodiment, before the intention determining module 1100, the method further includes: the word set construction module is used for acquiring commodity titles of a plurality of commodities, determining commodity entity words and entity types thereof in each commodity title, and constructing commodity entity word sets corresponding to a plurality of entity types; the training preparation module is used for acquiring dialogue texts input by a user in history as dialogue samples, labeling commodity entity words and entity types thereof related to purchase intention in the dialogue samples by adopting the commodity entity word sets, and constructing corresponding labeling sequences as supervision labels of the dialogue samples; and the first model training module is used for training the named entity model to a convergence state by adopting the dialogue sample and the supervision label thereof, so that the named entity model is suitable for identifying commodity entity words and entity types thereof in the dialogue text.

In a further embodiment, before the intention determining module 1100, the method further includes: the entity recognition module is used for acquiring a plurality of dialogue texts input by a user in a history way, and recognizing commodity entity words and entity types thereof in each dialogue text by adopting a preset named entity model; the second model training module is used for splicing commodity entity words and entity types thereof as compound words to replace original commodity entity words in corresponding dialogue texts, and training a text coding model to a convergence state by adopting the replaced dialogue texts so as to ensure that the text coding model is suitable for determining word vectors corresponding to each word element in the dialogue texts; the word cluster construction module is used for determining word vectors corresponding to compound words formed by splicing each commodity entity word and entity type thereof by adopting the text coding model trained until convergence, calculating the similarity between each compound word, and aggregating compound words with the similarity exceeding a preset threshold value to obtain a plurality of synonymous compound word clusters; and the word set forming module is used for classifying all the synonymous compound word clusters according to the entity types of compound words in the compound word clusters to obtain synonymous commodity entity words corresponding to the same entity types of the same genus to form the synonymous entity word set.

In a further embodiment, the intention judgment module 1100 includes: and the classification judging sub-module is used for acquiring the dialogue text input by the user, extracting the deep semantic information of the dialogue text by adopting a preset text classification model, acquiring corresponding semantic feature vectors for carrying out two classifications, and correspondingly determining whether the dialogue text has the purchase intention or not according to the classification result.

In a further embodiment, before the intention determining module 1100, the method further includes: the sample acquisition module is used for acquiring a single dialogue sample and a supervision label thereof from the prepared training set, wherein the supervision label characterizes whether the dialogue sample has a purchase intention or not; the classification module is used for extracting deep semantic information of the dialogue text by adopting a text classification model, obtaining corresponding semantic feature vectors for carrying out two classification, and obtaining corresponding classification results; and the iteration training module is used for determining a loss value of the classification result by adopting a supervision tag of the dialogue sample, updating the weight of the text classification model when the loss value does not reach a corresponding preset threshold value, and continuously calling other dialogue samples to perform iteration training until the text classification model converges.

In a further embodiment, the commodity searching module 1400 further includes: the scoring calculation module is used for obtaining the click rate and the conversion rate of each recommended commodity, matching the click rate and the conversion rate with corresponding weights, and calculating the recommendation score of each recommended commodity; and the commodity optimization module is used for screening the recommended commodity with the highest recommendation score as a high-quality commodity to carry out commodity recommendation.

In a further embodiment, the commodity searching module 1400 further includes: the sales crown determining module is used for obtaining sales volumes of the recommended commodities corresponding to each sales area and determining recommended commodities with the largest sales volumes corresponding to each sales area; and the region recommending module is used for acquiring the geographic position information of the user and determining recommended commodities with the largest sales volume in the sales region matched with the geographic position information to recommend the commodities.

In order to solve the technical problems, the embodiment of the application also provides computer equipment. As shown in fig. 8, the internal structure of the computer device is schematically shown. The computer device includes a processor, a computer readable storage medium, a memory, and a network interface connected by a system bus. The computer readable storage medium of the computer device stores an operating system, a database and computer readable instructions, the database can store a control information sequence, and the computer readable instructions can enable the processor to realize a commodity recommendation method when the computer readable instructions are executed by the processor. The processor of the computer device is used to provide computing and control capabilities, supporting the operation of the entire computer device. The memory of the computer device may have stored therein computer readable instructions that, when executed by the processor, cause the processor to perform the merchandise recommendation method of the present application. The network interface of the computer device is for communicating with a terminal connection. It will be appreciated by those skilled in the art that the structure shown in fig. 8 is merely a block diagram of some of the structures associated with the present application and is not limiting of the computer device to which the present application may be applied, and that a particular computer device may include more or fewer components than shown, or may combine certain components, or have a different arrangement of components.

The processor in this embodiment is configured to execute specific functions of each module and its sub-module in fig. 7, and the memory stores program codes and various data required for executing the above modules or sub-modules. The network interface is used for data transmission between the user terminal or the server. The memory in the present embodiment stores program codes and data necessary for executing all modules/sub-modules in the commodity recommendation apparatus of the present application, and the server can call the program codes and data of the server to execute the functions of all sub-modules.

The present application also provides a storage medium storing computer readable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the merchandise recommendation method of any of the embodiments of the present application.

Those skilled in the art will appreciate that implementing all or part of the above-described methods of embodiments of the present application may be accomplished by way of a computer program stored on a computer readable storage medium, which when executed, may comprise the steps of embodiments of the methods described above. The storage medium may be a computer readable storage medium such as a magnetic disk, an optical disk, a Read-On-y Memory (ROM), or a random access Memory (Random Access Memory, RAM).

In summary, the method and the device effectively solve the dilemma that the social electronic commerce scene lacks a convenient commodity browsing page, can provide convenient commodity recommendation reply suggestions for merchants, improve shopping guide service efficiency of the merchants, and promote commodity transaction.

Those of skill in the art will appreciate that the various operations, methods, steps in the flow, actions, schemes, and alternatives discussed in the present application may be alternated, altered, combined, or eliminated. Further, other steps, means, or steps in a process having various operations, methods, or procedures discussed in this application may be alternated, altered, rearranged, split, combined, or eliminated. Further, steps, measures, schemes in the prior art with various operations, methods, flows disclosed in the present application may also be alternated, altered, rearranged, decomposed, combined, or deleted.

The foregoing is only a partial embodiment of the present application, and it should be noted that, for a person skilled in the art, several improvements and modifications can be made without departing from the principle of the present application, and these improvements and modifications should also be considered as the protection scope of the present application.

Claims

1. The commodity recommending method is characterized by comprising the following steps of:

2. The merchandise recommendation method according to claim 1, further comprising the steps of, prior to obtaining the dialogue text entered by the user:

3. The merchandise recommendation method according to claim 1, further comprising the steps of, prior to obtaining the dialogue text entered by the user:

4. The merchandise recommendation method according to claim 1, wherein obtaining a dialogue text entered by a user, determining whether the dialogue text has a purchase intention, comprises:

5. The merchandise recommendation method according to claim 1, further comprising the steps of, prior to obtaining the dialogue text entered by the user:

6. The commodity recommending method according to claim 1, wherein the commodity entity word and the expanded entity word are spliced as search text, and after searching out the corresponding recommended commodity according to the search text, further comprising the steps of:

7. The commodity recommending method according to claim 1, wherein the commodity entity word and the expanded entity word are spliced as search text, and after searching out the corresponding recommended commodity according to the search text, further comprising the steps of:

8. A commodity recommendation device, comprising:

The intention judging module is used for acquiring dialogue texts input by a user and judging whether the dialogue texts have purchase intention or not;

the model identification module is used for identifying commodity entity words and entity types thereof in the dialogue text by adopting a preset named entity model when the dialogue text has purchase intention;

the synonym matching module is used for matching the expanded entity words synonymous with the commodity entity words from a preset synonym entity word set according to the entity types of the commodity entity words;

and the commodity searching module is used for splicing the commodity entity words and the expanded entity words to serve as search texts, and searching out corresponding recommended commodities according to the search texts.

9. A computer device comprising a central processor and a memory, characterized in that the central processor is arranged to invoke a computer program stored in the memory for performing the steps of the method according to any of claims 1 to 7.

10. A computer-readable storage medium, characterized in that it stores in the form of computer-readable instructions a computer program implemented according to the method of any one of claims 1 to 7, which, when invoked by a computer, performs the steps comprised by the corresponding method.