A search method

Abstract

The embodiment of the application discloses a search method, comprising: acquiring at least one current search word input by a current user and export enterprise information of the current user, determining at least one target extension word corresponding to each current search word and a similarity between each target extension word and the corresponding current search word, then performing search based on the at least one current search word and the at least one target extension word corresponding to each current search word to obtain a plurality of candidate search results, inputting procurement enterprise information and the export enterprise information in each candidate search result into a preset deep learning model to obtain an association degree between each candidate search result and the export enterprise information, then calculating a search score of each candidate search result according to the similarity and the association degree, and finally sorting the plurality of candidate search results according to the search score to obtain a target search result, so that the user can quickly and accurately search for a suitable purchaser for trade.

Description

Technical Field

[0001] This invention relates to the field of search technology, and more particularly to a search method. Background Technology

[0002] In the era of global trade, foreign trade opportunities are increasing, but it is becoming increasingly difficult for exporters to find suitable buyers (also known as suppliers) from a sea of ​​information. The main reasons include outdated or incorrect buyer information, mismatch between exporters and buyers' trade goods, and buyers' lack of purchasing intentions. Moreover, different exporters have different preferences for the purchase scale of different buyers. Therefore, in order to enable exporters to find suitable buyers and avoid wasting a lot of energy and costs, there is an urgent need for a search engine for finding buyers.

[0003] Currently, search engines used by exporters to find buyers typically generate search results by performing exact matching or word segmentation matching based on the search terms entered by the user (i.e., the exporter's salesperson). By calculating the relevance between the search results and the hit keywords, the search results are sorted and displayed from high to low relevance so that users can find suitable buyers for trade. However, this method only considers the search of surface information, making the search results merely superficial. In reality, it is difficult for users to find suitable buyers for trade based on the search results.

[0004] In addition, some search engines for buyers sort and display search results from high to low based on content popularity (e.g., recommended searches, clicks and views, click and view frequency). However, while this method may seem popular, most of the top search results are actually invalid, making it more difficult for users to find suitable buyers for trade based on the search results. Summary of the Invention

[0005] Therefore, it is necessary to propose a search method to address the above problems, enabling users to quickly and accurately find suitable buyers for trade.

[0006] To achieve the above objectives, the present invention provides a search method, the method comprising:

[0007] Obtain at least one current search term entered by the current user and the exporting company information of the current user;

[0008] Determine at least one target expanded term corresponding to each current search term and the similarity between each target expanded term and its corresponding current search term;

[0009] Based on at least one current search term and at least one target extended term corresponding to each current search term, multiple candidate search results are obtained. The purchasing enterprise information and the exporting enterprise information in each candidate search result are input into a preset deep learning model to obtain the correlation between each candidate search result and the exporting enterprise information.

[0010] The search score for each candidate search result is calculated based on the similarity between each target expanded term and the corresponding current search term and the relevance between each candidate search result and the export enterprise information.

[0011] The target search result is obtained by sorting multiple candidate search results according to the search score of each candidate search result.

[0012] Optionally, before sorting multiple candidate search results according to the search score of each candidate search result to obtain the target search result, the method further includes:

[0013] Determine the relevance compensation amount between each target expanded term and its corresponding current search term;

[0014] The search score for each candidate search result is calculated based on the similarity between each target expanded term and its corresponding current search term, the relevance compensation between each target expanded term and its corresponding current search term, and the correlation between each candidate search result and the export enterprise information.

[0015] Optionally, before sorting multiple candidate search results according to the search score of each candidate search result to obtain the target search result, the method further includes:

[0016] Determine the relevance score of each target expanded term to the corresponding current search term in each candidate search result;

[0017] The search score for each candidate search result is calculated based on the similarity between each target extended term and the corresponding current search term, the relevance compensation between each target extended term and the corresponding current search term, the correlation between each candidate search result and the export enterprise information, and the relevance score of each target extended term and the corresponding current search term in each candidate search result.

[0018] Optionally, the step of calculating the search score for each candidate search result based on the similarity between each target extended term and the corresponding current search term, the relevance compensation amount between each target extended term and the corresponding current search term, the correlation between each candidate search result and the export enterprise information, and the relevance score of each target extended term and the corresponding current search term in each candidate search result includes:

[0019] The search score for each candidate search result is calculated using the following formula:

[0020]

[0021]

[0022] Wherein, Score(q) is the search score of the q-th candidate search result, R(q) is the correlation between the q-th candidate search result and the export enterprise information, m is the total number of at least one current search term entered by the current user, n is the total number of at least one target extended term corresponding to the i-th current search term, Searchwords(i) is the total value of the i-th current search term and the n target extended terms corresponding to the i-th current search term, SIM_keyword(j) is the similarity between the j-th target extended term and the corresponding i-th current search term, SearchScore(j) is the relevance score between the j-th target extended term and the corresponding i-th current search term, and C(j) is the relevance compensation amount between the j-th target extended term and the corresponding i-th current search term.

[0023] Optionally, determining at least one target expanded term corresponding to each current search term and the similarity between each target expanded term and the corresponding current search term includes:

[0024] Identify at least one candidate expanded term for each current search term;

[0025] Each current search term is sequentially input into the preset Word2vec model to obtain the first word vector of each current search term, and each candidate expanded term is input into the preset Word2vec model to obtain the second word vector of each candidate expanded term;

[0026] Determine the similarity between the second word vector of each candidate extended word and the first word vector of the corresponding current search word, and take the candidate extended words with similarity greater than or equal to the similarity threshold as the target extended words of the corresponding current search word, and take the similarity as the similarity between the target extended word and the corresponding current search word.

[0027] Optionally, determining the relevance compensation amount between each target expanded term and its corresponding current search term includes:

[0028] If each target expanded term and its corresponding current search term have been searched by historical users, the historical users who have searched each target expanded term and its corresponding current search term will be considered as target users.

[0029] Obtain enterprise information for all target users;

[0030] By combining any two target users into a target user group, multiple target user groups can be obtained.

[0031] Calculate the common dimensions of user information between two target users in each target user group to obtain multiple common dimensions;

[0032] Calculate a weighted average of multiple terms with the same dimensions, and use the weighted average as the relevance compensation between each target extended term and the corresponding current search term.

[0033] Optionally, determining the relevance compensation amount between each target expanded term and its corresponding current search term includes:

[0034] If each target expanded term and its corresponding current search term have not been searched by historical users, initialize the relevance compensation amount between each target expanded term and its corresponding current search term according to the threshold range.

[0035] Optionally, determining the relevance score between each target expanded term and its corresponding current search term in each candidate search result includes:

[0036] Based on the TF / IDF model, the word frequency of each target expanded word and its corresponding current search term in each candidate search result is calculated, and the word frequency is used as the relevance score of each target expanded word and its corresponding current search term in each candidate search result.

[0037] Optionally, before obtaining multiple candidate search results by searching based on at least one current search term and at least one target extended term corresponding to each current search term, and inputting the purchasing company information and the exporting company information from each candidate search result into a preset deep learning model to obtain the correlation between each candidate search result and the exporting company information, the method further includes:

[0038] Collect historical export enterprise information, historical purchase enterprise information, and the historical correlation between the historical export enterprise information and the historical purchase enterprise information;

[0039] The historical export enterprise information, the historical purchase enterprise information, and the historical correlation degree are divided into training datasets and validation datasets according to a preset ratio.

[0040] The training dataset is input into the initial deep learning model for training to obtain the basic deep learning model;

[0041] The validation dataset is input into the base deep learning model to obtain the predicted correlation.

[0042] If the predicted correlation is greater than or equal to the correlation threshold, the basic deep learning model is used as the preset deep learning model.

[0043] Optionally, the method further includes:

[0044] If the predicted correlation is less than the correlation threshold, the training dataset is continued to be input into the basic deep learning model for further training until the validation dataset is input into the retrained basic deep learning model to obtain the retrained predicted correlation. If the retrained predicted correlation is greater than or equal to the correlation threshold, the retrained basic deep learning model is used as the preset deep learning model.

[0045] The present invention provides the following advantages: The method obtains at least one current search term input by the current user and the current user's export company information, determines at least one target extended term corresponding to each current search term, and determines the similarity between each target extended term and the corresponding current search term. Then, based on at least one current search term and at least one target extended term corresponding to each current search term, it performs a search to obtain multiple candidate search results. The purchasing company information and export company information in each candidate search result are input into a preset deep learning model to obtain the correlation between each candidate search result and the export company information. Next, based on the similarity between each target extended term and the corresponding current search term and the correlation between each candidate search result and the export company information, a search score for each candidate search result is calculated. Finally, based on the search scores of each candidate search result, the multiple candidate search results are sorted to obtain the target search result. This allows for a search that simultaneously considers the needs of both the buyer and the user when searching for a suitable buyer. This method optimizes the sorting of multiple candidate search results, greatly improving the effective use of the search engine and the user's acceptance of the search engine. This enables users to quickly and accurately find suitable buyers for trade, significantly improving the efficiency of the search engine and promoting the development of foreign trade. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] in:

[0048] Figure 1This is a schematic diagram of a search method in an embodiment of this application;

[0049] Figure 2 This is a schematic diagram of a search device according to an embodiment of this application;

[0050] Figure 3 Internal structural diagrams of computer devices in some embodiments are shown. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Please see Figure 1 This is a schematic diagram of a search method according to an embodiment of this application. The method includes:

[0053] Step 110: Obtain at least one current search term entered by the current user and the current user's export company information.

[0054] It should be noted that the current user can be a salesperson of an exporter. Therefore, the current user has export enterprise information corresponding to their company. This enterprise information includes, but is not limited to, the products the company operates in, the industry it operates in, the main export regions the company exports to, the supported export freight methods, and historical or typical trade records. In some scenarios, this enterprise information may also include the current user's user information, that is, the user information edited by the current user themselves.

[0055] Step 120: Determine at least one target expanded term corresponding to each current search term and the similarity between each target expanded term and the corresponding current search term.

[0056] In some embodiments, after obtaining at least one current search term, at least one word related to the current search term can be determined based on each current search term, such as a synonym. Then, each synonym is compared with the current search term to obtain the similarity between each synonym and the current search term. If the similarity exceeds a certain threshold, the synonym is used as the target expanded word of the current search term. Alternatively, the similarity can also be used as the similarity between the corresponding target expanded word and the current search term. It is understood that by obtaining at least one word related to the current search term for each current search term, the search term can be effectively expanded, avoiding missed searches. By comparing the similarity between each synonym and the current search term, abnormal synonyms can be effectively eliminated, avoiding invalid expansions and invalid searches.

[0057] It should be noted that at least one word related to the current search term can include the current search term itself. Therefore, each current search term must have at least one synonym, meaning that at least one synonym must be the current search term. Furthermore, the similarity between the current search term and its synonym must meet a certain threshold. Thus, each current search term must have at least one target extended term.

[0058] Step 130: Based on at least one current search term and at least one target extended term corresponding to each current search term, perform a search to obtain multiple candidate search results. Input the purchasing enterprise information and exporting enterprise information in each candidate search result into a preset deep learning model to obtain the correlation between each candidate search result and the exporting enterprise information.

[0059] Each candidate search result represents relevant information about a buyer. The correlation can be the predicted transaction completion rate between the buyer and exporter in the candidate search results. This correlation helps users quickly and accurately find suitable buyers for trade. The buyer information may include, but is not limited to, industry sector, products and series, import freight method, location, import source region, import scale in the past 3 years, import frequency in the past 3 years, and recent trade information.

[0060] It should be noted that the preset deep learning model is a pre-trained deep learning model. Specifically, it can be obtained by training historical information on purchasing companies, exporting companies, and the historical correlation between historical information on purchasing companies and exporting companies (i.e., historical transaction completion status).

[0061] In some embodiments, multiple candidate search results can be obtained by searching with at least one current search term and at least one target extended term corresponding to each current search term. Then, the purchasing enterprise information and exporting enterprise information in each candidate search result are input into a preset deep learning model to obtain the correlation between each candidate search result and the exporting enterprise information.

[0062] Step 140: Calculate the search score for each candidate search result based on the similarity between each target expanded term and the corresponding current search term and the relevance between each candidate search result and the export enterprise information.

[0063] In some embodiments, a first sum value between each current search term and at least one corresponding target extended term can be obtained based on the similarity between each target extended term and the corresponding current search term. Then, a second sum value of at least one current search term can be obtained based on the first sum value between each current search term and the corresponding at least one target extended term. Finally, a search score for each candidate search structure can be calculated based on the second sum value of the at least one current search term and the correlation between each candidate search result and the export enterprise information.

[0064] Step 150: Sort the multiple candidate search results according to the search score of each candidate search result to obtain the target search result.

[0065] In some embodiments, multiple candidate search results can be sorted from high search score to low search score according to the search score of each candidate search result, so as to obtain multiple sorted search results, i.e. target search results.

[0066] In this embodiment, by acquiring at least one current search term input by the current user and the current user's export company information, and determining at least one target extended term corresponding to each current search term and the similarity between each target extended term and the corresponding current search term, a search is performed based on at least one current search term and at least one target extended term corresponding to each current search term to obtain multiple candidate search results. The purchasing company information and export company information in each candidate search result are input into a preset deep learning model to obtain the correlation between each candidate search result and the export company information. Then, the search score of each candidate search result is calculated based on the similarity between each target extended term and the corresponding current search term and the correlation between each candidate search result and the export company information. Finally, the multiple candidate search results are sorted according to the search scores of each candidate search result to obtain the target search result. This allows for a search that simultaneously considers the needs of both the buyer and the user when searching for a suitable buyer. This method optimizes the sorting of multiple candidate search results, which can greatly improve the effective use of the search engine and the user's acceptance of the search engine. As a result, users can quickly and accurately search for suitable buyers for trade, greatly improving the efficiency of the search engine and promoting the development of foreign trade transactions.

[0067] In one feasible implementation, before step 150 in the above embodiment, which sorts multiple candidate search results according to the search score of each candidate search result to obtain the target search result, the method further includes: determining the relevance compensation amount between each target extended term and the corresponding current search term; and calculating the search score of each candidate search result based on the similarity between each target extended term and the corresponding current search term, the relevance compensation amount between each target extended term and the corresponding current search term, and the correlation between each candidate search result and the export enterprise information.

[0068] It should be noted that the relevance compensation amount between each target extended term and its corresponding current search term can be the percentage of commonalities in user information among historical users who have searched for each target extended term and its corresponding current search term. That is, the relevance compensation amount between each target extended term and its corresponding current search term is used as a reference to the percentage of commonalities in user information among historical users who have frequently searched for each target extended term and its corresponding current search term. In addition, if no historical users have searched for each target extended term and its corresponding current search term, the relevance compensation amount between each target extended term and its corresponding current search term is set to the default value.

[0069] In some embodiments, a first sum value between each current search term and at least one corresponding target extended term can be obtained based on the similarity between each target extended term and the corresponding current search term. Then, a second sum value for at least one current search term can be obtained based on the first sum value. Next, a third sum value between each current search term and at least one corresponding target extended term can be obtained based on the relevance compensation amount between each target extended term and the corresponding current search term. Finally, a fourth sum value for at least one current search term can be obtained based on the third sum value. Finally, a search score for each candidate search structure can be calculated based on the second sum value, the fourth sum value, and the relevance between each candidate search result and the export enterprise information. This allows multiple candidate search results to be sorted according to their search scores to obtain the target search result.

[0070] In this embodiment, by determining the relevance compensation amount between each target extended term and the corresponding current search term, that is, by referring to the proportion of similarities in user information between users who have frequently searched each target extended term and the corresponding current search term in the past, the search score of each candidate search result can be improved. Then, the multiple candidate search results are sorted according to the search score of each candidate search result to obtain the target search result. This allows users to find suitable buyers for trade more quickly and accurately, which further greatly improves the efficiency of the search engine and promotes the development of foreign trade transactions.

[0071] In one feasible implementation, before step 150 in the above embodiment, which sorts multiple candidate search results according to the search score of each candidate search result to obtain the target search result, the method further includes: determining the relevance score of each target extended term and its corresponding current search term in each candidate search result; and calculating the search score of each candidate search result based on the similarity between each target extended term and its corresponding current search term, the relevance compensation amount between each target extended term and its corresponding current search term, the correlation between each candidate search result and the export enterprise information, and the relevance score of each target extended term and its corresponding current search term in each candidate search result.

[0072] It should be noted that the relevance score of each target expanded term and its corresponding current search term in each candidate search result can be the word frequency of each target expanded term and its corresponding current search term in each candidate search result.

[0073] In some embodiments, a fifth sum value between each current search term and at least one corresponding target extended term can be obtained based on the relevance score between each target extended term and the corresponding current search term in each candidate search result; a sixth sum value for at least one current search term can be obtained based on the fifth sum value between each current search term and at least one corresponding target extended term; a first sum value between each current search term and at least one corresponding target extended term can be obtained based on the similarity between each target extended term and the corresponding current search term; a second sum value for at least one current search term can be obtained based on the first sum value between each current search term and at least one corresponding target extended term; a third sum value between each current search term and at least one corresponding target extended term can be obtained based on the relevance compensation amount between each target extended term and the corresponding current search term; a fourth sum value for at least one current search term can be obtained based on the third sum value between each current search term and at least one corresponding target extended term; finally, a search score for each candidate search structure can be calculated based on the second sum value, the fourth sum value, the sixth sum value, and the correlation between each candidate search result and the export enterprise information; thus, multiple candidate search results can be sorted based on the search score of each candidate search result to obtain the target search result.

[0074] In this embodiment, the relevance score of each target extended term and its corresponding current search term in each candidate search result is obtained by using the word frequency of each target extended term and its corresponding current search term in each candidate search result. The search score of each candidate search result is then used to sort the multiple candidate search results to obtain the target search result. This allows users to find suitable buyers for trade more quickly and accurately, further greatly improving the efficiency of the search engine and promoting the development of foreign trade transactions.

[0075] In one feasible implementation, the search score of each candidate search result in the above embodiments is calculated based on the similarity between each target extended term and the corresponding current search term, the relevance compensation amount between each target extended term and the corresponding current search term, the correlation between each candidate search result and the export enterprise information, and the relevance score of each target extended term and the corresponding current search term in each candidate search result, including:

[0076] The search score for each candidate search result is calculated using the following formula:

[0077]

[0078]

[0079] Wherein, Score(q) is the search score of the q-th candidate search result, R(q) is the correlation between the q-th candidate search result and the export enterprise information, m is the total number of at least one current search term entered by the current user, n is the total number of at least one target extended term corresponding to the ith current search term, Searchwords(i) is the total value of the ith current search term and the n target extended terms corresponding to the ith current search term, SIM_keyword(j) is the similarity between the j-th target extended term and the corresponding ith current search term, SearchScore(j) is the relevance score between the j-th target extended term and the corresponding ith current search term, and C(j) is the relevance compensation amount between the j-th target extended term and the corresponding ith current search term.

[0080] In this embodiment of the application, the preferred formula for calculating the search score of each candidate search result is shown, which makes it easier for operators to use the above formula to calculate the search score of each candidate search result, thereby optimizing the ranking of multiple candidate search results. This greatly improves the effective use of the search engine and the user's acceptance of the search engine, enabling users to quickly and accurately find suitable buyers for trade, greatly improving the efficiency of the search engine and promoting the development of foreign trade transactions.

[0081] In one feasible implementation, step 120 in the above embodiment, determining at least one target extended word corresponding to each current search term and the similarity between each target extended word and the corresponding current search term, includes: determining at least one candidate extended word corresponding to each current search term; sequentially inputting each current search term into a preset Word2vec model to obtain a first word vector of each current search term, and inputting each candidate extended word into the preset Word2vec model to obtain a second word vector of each candidate extended word; determining the similarity between the second word vector of each candidate extended word and the first word vector of the corresponding current search term, and taking the candidate extended word with a similarity greater than or equal to a similarity threshold as the target extended word of the corresponding current search term, and taking the similarity as the similarity between the target extended word and the corresponding current search term.

[0082] The similarity threshold is a critical value for similarity obtained by the operator based on a large number of experiments or statistics. This similarity threshold is generally set to 0.8, but it can also be adjusted according to the actual needs of the operator.

[0083] It should be noted that the Word2vec model is a model that can calculate word vectors. In this application, the preset Word2vec model is trained by the operator using product encyclopedia information and / or trade product introduction information. It can be understood that since the transaction between the user (i.e., the exporter's salesperson) and the search results (buyer) in this application is for products, and the user input is generally also product-related search terms, this application uses product-related information to train the preset Word2vec model.

[0084] Furthermore, the similarity between the second word vector of each candidate expanded word and the first word vector of the corresponding current search word can be the cosine similarity between the second word vector of each candidate expanded word and the first word vector of the corresponding current search word, that is, the cosine value between the first word vector and the second word vector.

[0085] In some embodiments, at least one candidate extended word corresponding to each current search term may be at least one synonym (or similar word, etc.) corresponding to each current search term, and the at least one synonym corresponding to the current search term may include each current search term itself. Therefore, each current search term has at least one candidate extended word, that is, there is a candidate extended word that is definitely the current search term, and the similarity between the current search term and the candidate extended word when it is the current search term is greater than or equal to the similarity threshold. Therefore, each current search term must have at least one target extended word.

[0086] In another feasible implementation, after determining at least one candidate extended word corresponding to each current search term, each current search term and at least one candidate search term corresponding to each current search term are sequentially input into a preset Word2vec model to obtain the first word vector of each current search term and the second word vector of each candidate extended word. Then, the similarity between the second word vector of each candidate extended word and the first word vector of the corresponding current search term is determined, and the candidate extended word with a similarity greater than or equal to the similarity threshold is taken as the target extended word of the corresponding current search term, and the similarity is taken as the similarity between the target extended word and the corresponding current search term.

[0087] In this embodiment, by determining at least one candidate extended word corresponding to each current search term, the current search term can be effectively expanded, avoiding missed searches. By sequentially inputting each current search term into a preset Word2vec model to obtain the first word vector of each current search term, and inputting each candidate extended word into the preset Word2vec model to obtain the second word vector of each candidate extended word, the similarity between the second word vector of each candidate extended word and the first word vector of the corresponding current search term is determined. Candidate extended words with similarity greater than or equal to the similarity threshold are taken as target extended words of the corresponding current search term, and the similarity is taken as the similarity between the target extended word and the corresponding current search term. This effectively eliminates abnormal candidate extended words, avoids the expansion of invalid words, and effectively avoids finding invalid candidate search results.

[0088] In one feasible implementation, determining the relevance compensation amount between each target extended term and its corresponding current search term in the above embodiments includes: if each target extended term and its corresponding current search term have been searched by historical users, taking the historical users who have searched each target extended term and its corresponding current search term as target users; obtaining the enterprise information of all target users; forming target user groups from any two target users to obtain multiple target user groups; calculating the common dimensions of user information between two target users in each target user group to obtain multiple common dimensions; calculating the weighted average of the multiple common dimensions, and using the weighted average as the relevance compensation amount between each target extended term and its corresponding current search term.

[0089] It should be noted that, in addition to exporting company information, target users can also edit their own user information, which may include, but is not limited to, industry sector, products, series, export freight method, export scale, export region, customer type, etc.

[0090] In some embodiments, when calculating the common dimensions of user information between two target users in each target user group, multiple common dimensions are obtained. In the case where the user information is industry field, product, series, export freight method, export region, or customer type, the calculation method of the common dimension is as follows: In a target user group, there are target user A and target user B. The user information between target user A and target user B has the same industry field and export region, that is, the common points of user information account for 2 / 6 of the total user information. Then, the common dimension of user information between target user A and target user B is 1 / 3.

[0091] In this embodiment, when each target extended term and its corresponding current search term have been searched by historical users, the historical users who have searched each target extended term and its corresponding current search term are taken as target users. The same dimension between any two target users is calculated, and the weighted average of multiple same dimensions is calculated. The weighted average is used as the relevance compensation amount between each target extended term and its corresponding current search term. This allows for a higher search score for each candidate search result. The multiple candidate search results are then sorted according to the search score of each candidate search result to obtain the target search result. This enables users to find suitable buyers for trade more quickly and accurately, further greatly improving the efficiency of the search engine and promoting the development of foreign trade transactions.

[0092] In one feasible implementation, determining the relevance compensation amount between each target extended term and the corresponding current search term in the above embodiments includes: when each target extended term and the corresponding current search term have not been searched by historical users, initializing the relevance compensation amount between each target extended term and the corresponding current search term according to a threshold range.

[0093] It should be noted that the threshold range is an initial relevance compensation range obtained by the operator based on a large number of experiments or statistics. It can be understood that when the search engine is initially used, each target extended term and its corresponding current search term have not been searched by historical users. Therefore, there is no target user reference. At this time, the relevance compensation amount between each target extended term and its corresponding current search term can be initialized according to the threshold range. In this application, the threshold range can be set to [0.1, 0.2]. Of course, it can also be adjusted according to the actual needs of the operator.

[0094] In this embodiment, when each target extended term and its corresponding current search term have not been searched by historical users (i.e., when the search engine is initially used and no target users exist), the relevance compensation amount between each target extended term and its corresponding current search term is initialized according to a threshold range. This cold-start random initial assignment of a decimal value to the relevance compensation amount between each target extended term and its corresponding current search term allows for a higher search score for each candidate search result. The multiple candidate search results are then sorted based on their search scores to obtain the target search result. This enables users to find suitable buyers for trade more quickly and accurately, further greatly improving the efficiency of the search engine and promoting the development of foreign trade.

[0095] In one feasible implementation, determining the relevance score of each target extended term and its corresponding current search term in each candidate search result in the above embodiments includes: calculating the word frequency of each target extended term and its corresponding current search term in each candidate search result based on the TF / IDF model, and using the word frequency as the relevance score of each target extended term and its corresponding current search term in each candidate search result.

[0096] Among them, TF / IDF (Term Frequency / Inverse Document Frequency) is a commonly used weighting technique for information retrieval and data mining. TF stands for Term Frequency, and IDF stands for Inverse Document Frequency.

[0097] In some embodiments, the word frequency of each target expanded term and its corresponding current search term in each candidate search result can be calculated based on the TF / IDF model, and the word frequency can be used as the relevance score of each target expanded term and its corresponding current search term in each candidate search result.

[0098] In this embodiment, by using the TF / IDF model, the word frequency of each target extended term and its corresponding current search term in each candidate search result is calculated, thereby obtaining the relevance score of each target extended term and its corresponding current search term in each candidate search result. This allows for the calculation of the search score of each candidate search result. Then, based on the search score of each candidate search result, multiple candidate search results are sorted to obtain the target search result. This enables users to find suitable buyers for trade more quickly and accurately, further greatly improving the efficiency of the search engine and promoting the development of foreign trade transactions.

[0099] In one feasible implementation, before step 130 in the above embodiments, which involves searching for multiple candidate search results based on at least one current search term and at least one target extended term corresponding to each current search term, and inputting the purchasing enterprise information and export enterprise information in each candidate search result into a preset deep learning model to obtain the correlation between each candidate search result and the export enterprise information, the method further includes: collecting historical export enterprise information, historical purchasing enterprise information, and historical correlation between historical export enterprise information and historical purchasing enterprise information; dividing the historical export enterprise information, historical purchasing enterprise information, and historical correlation into a training dataset and a validation dataset according to a preset ratio; inputting the training dataset into an initial deep learning model for training to obtain a basic deep learning model; inputting the validation dataset into the basic deep learning model to obtain a predicted correlation; and using the basic deep learning model as the preset deep learning model when the predicted correlation is greater than or equal to a correlation threshold.

[0100] The preset ratio and correlation threshold are obtained by the operator based on a large number of experiments or statistics. In some embodiments, the preset ratio can be 4:1 and the correlation threshold can be 0.8.

[0101] It should be noted that historical export enterprise information includes, but is not limited to, the enterprise's products, industry, main export regions, supported export freight methods, and historical or typical trade information. Historical purchase enterprise information includes, but is not limited to, industry sector, products and series, import freight methods, location, import source region, import volume in the past 3 years, import frequency in the past 3 years, and recent trade information. The historical correlation between historical export enterprise information and historical purchase enterprise information is the historical transaction completion status between them. For example, in some embodiments, if a transaction occurs between historical export enterprise information and historical purchase enterprise information, the historical transaction completion status is 1, i.e., the historical correlation is 1; if no transaction occurs between historical export enterprise information and historical purchase enterprise information, the historical transaction completion status is 0, i.e., the historical correlation is 0.

[0102] It should also be noted that the training dataset is used for model training, so it includes not only historical exporter information and historical purchaser information, but also the historical correlation between them. The validation dataset is only used for validation, so it can include only historical exporter information and historical purchaser information, without including the historical correlation between them.

[0103] In this embodiment, the training set of the deep learning model is validated by collecting historical export enterprise information, historical purchase enterprise information, and the historical correlation between historical export enterprise information and historical purchase enterprise information. When the predicted correlation obtained by the validation is greater than or equal to the correlation threshold, the basic deep learning model is used as the preset deep learning model so that the probability of a transaction between the current user and the purchaser in the candidate search results can be predicted. The search score of each candidate search result can be obtained, and the multiple candidate search results are sorted according to the search score of each candidate search result to obtain the target search result. This allows users to find suitable purchasers for trade more quickly and accurately, which further greatly improves the efficiency of the search engine and promotes the development of foreign trade transactions.

[0104] In one feasible implementation, the method in the above embodiments further includes: if the predicted correlation degree is less than the correlation degree threshold, continuing to input the training dataset into the basic deep learning model for training until the verification dataset is input into the retrained basic deep learning model to obtain the retrained predicted correlation degree, and if the retrained predicted correlation degree is greater than or equal to the correlation degree threshold, using the retrained basic deep learning model as the preset deep learning model.

[0105] In this embodiment, when the predicted relevance is less than the relevance threshold (i.e., when the prediction rate of the basic deep learning model is low), the basic deep learning model is retrained and validated until the predicted relevance after retraining is greater than or equal to the relevance threshold. The retrained basic deep learning model is then used as the preset deep learning model. This ensures the accuracy of predicting the probability of a transaction between the current user and the buyers in the candidate search results. It also allows for the calculation of the search score for each candidate search result. By ranking multiple candidate search results based on their search scores, the target search result is obtained. This enables users to find suitable buyers for trade more quickly and accurately, further greatly improving the efficiency of the search engine and promoting the development of foreign trade.

[0106] In some embodiments, this application also provides a search device. See also Figure 2 This is a schematic diagram of a search device according to an embodiment of this application. The device 210 includes:

[0107] The acquisition module 211 is used to acquire at least one current search term entered by the current user and the current user's export enterprise information;

[0108] The similarity determination module 212 is used to determine at least one target extended word corresponding to each current search word and the similarity between each target extended word and the corresponding current search word;

[0109] The correlation determination module 213 is used to search for multiple candidate search results based on at least one current search term and at least one target extended term corresponding to each current search term, and input the purchasing enterprise information and export enterprise information in each candidate search result into a preset deep learning model to obtain the correlation between each candidate search result and the export enterprise information.

[0110] The search score determination module 214 is used to calculate the search score of each candidate search result based on the similarity between each target expanded term and the corresponding current search term and the relevance between each candidate search result and the export enterprise information.

[0111] The sorting module 215 is used to sort multiple candidate search results according to the search score of each candidate search result to obtain the target search result.

[0112] In this embodiment of the application, the relevant contents of the above-mentioned acquisition module 211, similarity determination module 212, relevance determination module 213, search score determination module 214, and ranking module 215 can be found in the following references. Figure 1 The contents of the illustrated embodiments will not be repeated here.

[0113] In this embodiment, by acquiring at least one current search term input by the current user and the current user's export company information, and determining at least one target extended term corresponding to each current search term and the similarity between each target extended term and the corresponding current search term, a search is performed based on at least one current search term and at least one target extended term corresponding to each current search term to obtain multiple candidate search results. The purchasing company information and export company information in each candidate search result are input into a preset deep learning model to obtain the correlation between each candidate search result and the export company information. Then, the search score of each candidate search result is calculated based on the similarity between each target extended term and the corresponding current search term and the correlation between each candidate search result and the export company information. Finally, the multiple candidate search results are sorted according to the search scores of each candidate search result to obtain the target search result. This allows for a search that simultaneously considers the needs of both the buyer and the user when searching for a suitable buyer. This method optimizes the sorting of multiple candidate search results, which can greatly improve the effective use of the search engine and the user's acceptance of the search engine. As a result, users can quickly and accurately search for suitable buyers for trade, greatly improving the efficiency of the search engine and promoting the development of foreign trade transactions.

[0114] In some embodiments, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform a search method in the above-described method embodiments.

[0115] In some embodiments, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs a search method according to the above method embodiments.

[0116] Figure 3 The diagram illustrates the internal structure of a computer device in some embodiments. This computer device may specifically be a terminal, a server, or a gateway. Figure 3 As shown, the computer device includes a processor, memory, and network interface connected via a system bus.

[0117] The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program. When executed by a processor, this computer program causes the processor to perform the steps in the above method embodiments. The internal memory may also store a computer program, which, when executed by a processor, causes the processor to perform the steps in the above method embodiments. Those skilled in the art will understand that... Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0118] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods.

[0119] Any references to memory, storage, database, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0120] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0121] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A search method, characterized in that the method comprises: Obtain at least one current search term entered by the current user and the exporting company information of the current user; Determine at least one target expanded term corresponding to each current search term and the similarity between each target expanded term and its corresponding current search term; Multiple candidate search results are obtained by searching based on at least one current search term and at least one target extended term corresponding to each current search term. The purchasing enterprise information and the exporting enterprise information in each candidate search result are input into a preset deep learning model to obtain the correlation degree between each candidate search result and the exporting enterprise information. The correlation degree is the predicted transaction completion status between the purchasing enterprise and the exporting enterprise. The preset deep learning model is obtained by training historical purchasing enterprise information, exporting enterprise information, and historical transaction completion status between historical purchasing enterprise information and exporting enterprise information. The search score for each candidate search result is calculated based on the similarity between each target expanded term and the corresponding current search term and the relevance between each candidate search result and the export enterprise information. The target search result is obtained by sorting multiple candidate search results according to the search score of each candidate search result.

2. The method according to claim 1, characterized in that, Before sorting multiple candidate search results according to the search score of each candidate search result to obtain the target search result, the method further includes: Determine the relevance compensation amount between each target expanded term and its corresponding current search term; The search score for each candidate search result is calculated based on the similarity between each target expanded term and its corresponding current search term, the relevance compensation between each target expanded term and its corresponding current search term, and the correlation between each candidate search result and the export enterprise information.

3. The method according to claim 2, characterized in that, Before sorting multiple candidate search results according to the search score of each candidate search result to obtain the target search result, the method further includes: Determine the relevance score of each target expanded term to the corresponding current search term in each candidate search result; The search score for each candidate search result is calculated based on the similarity between each target extended term and the corresponding current search term, the relevance compensation between each target extended term and the corresponding current search term, the correlation between each candidate search result and the export enterprise information, and the relevance score of each target extended term and the corresponding current search term in each candidate search result.

4. The method according to claim 3, characterized in that, The step of calculating the search score for each candidate search result based on the similarity between each target expanded term and its corresponding current search term, the relevance compensation amount between each target expanded term and its corresponding current search term, the correlation between each candidate search result and the export enterprise information, and the relevance score of each target expanded term and its corresponding current search term in each candidate search result includes: The search score for each candidate search result is calculated using the following formula: ； ； in, For the first The search score of each candidate search result. For the first The correlation between each candidate search result and the exporting company information. The total number of at least one current search term entered by the current user. For the first The total number of at least one target expanded term corresponding to each current search term. For the first The current search term and the first The current search term corresponds to The total value of each target expanded word, For the first The target expanded word and its corresponding first Similarity between current search terms For the first The target expanded word and its corresponding first Relevance score between current search terms For the first The target expanded word and its corresponding first The amount of relevance compensation between current search terms.

5. The method according to claim 1, characterized in that, Determining at least one target expanded term corresponding to each current search term and the similarity between each target expanded term and the corresponding current search term includes: Identify at least one candidate expanded term for each current search term; Each current search term is sequentially input into the preset Word2vec model to obtain the first word vector of each current search term, and each candidate expanded term is input into the preset Word2vec model to obtain the second word vector of each candidate expanded term; Determine the similarity between the second word vector of each candidate extended word and the first word vector of the corresponding current search word, and take the candidate extended words with similarity greater than or equal to the similarity threshold as the target extended words of the corresponding current search word, and take the similarity as the similarity between the target extended word and the corresponding current search word.

6. The method according to claim 2, characterized in that, The determination of the relevance compensation amount between each target expanded term and the corresponding current search term includes: If each target expanded term and its corresponding current search term have been searched by historical users, the historical users who have searched each target expanded term and its corresponding current search term will be considered as target users. Obtain enterprise information for all target users; By combining any two target users into a target user group, multiple target user groups can be obtained. Calculate the common dimensions of user information between two target users in each target user group to obtain multiple common dimensions; Calculate a weighted average of multiple terms with the same dimensions, and use the weighted average as the relevance compensation between each target extended term and the corresponding current search term.

7. The method according to claim 2, characterized in that, The determination of the relevance compensation amount between each target expanded term and the corresponding current search term includes: If each target expanded term and its corresponding current search term have not been searched by historical users, initialize the relevance compensation amount between each target expanded term and its corresponding current search term according to the threshold range.

8. The method according to claim 3, characterized in that, Determining the relevance score between each target expanded term and its corresponding current search term in each candidate search result includes: Based on the TF / IDF model, the word frequency of each target expanded word and its corresponding current search term in each candidate search result is calculated, and the word frequency is used as the relevance score of each target expanded word and its corresponding current search term in each candidate search result.

9. The method according to claim 1, characterized in that, Before obtaining multiple candidate search results by searching based on at least one current search term and at least one target extended term corresponding to each current search term, and inputting the purchasing company information and the exporting company information from each candidate search result into a preset deep learning model to obtain the correlation between each candidate search result and the exporting company information, the method further includes: Collect historical export enterprise information, historical purchase enterprise information, and the historical correlation between the historical export enterprise information and the historical purchase enterprise information; The historical export enterprise information, the historical purchase enterprise information, and the historical correlation degree are divided into training datasets and validation datasets according to a preset ratio. The training dataset is input into the initial deep learning model for training to obtain the basic deep learning model; The validation dataset is input into the base deep learning model to obtain the predicted correlation. If the predicted correlation is greater than or equal to the correlation threshold, the basic deep learning model is used as the preset deep learning model.

10. The method according to claim 9, characterized in that, The method further includes: If the predicted correlation is less than the correlation threshold, the training dataset is continued to be input into the basic deep learning model for further training until the validation dataset is input into the retrained basic deep learning model to obtain the retrained predicted correlation. If the retrained predicted correlation is greater than or equal to the correlation threshold, the retrained basic deep learning model is used as the preset deep learning model.

Images