Query word processing method and device, electronic equipment and readable storage medium

Abstract

Embodiments of the present application provide a query word processing method and device, electronic equipment and readable storage medium, in which the original query word adapted to the user to be recommended is obtained, and the original query word is adjusted according to the auxiliary feature to generate the target query word. The auxiliary feature includes the object feature of the network object hit by the original query word, the user attribute feature of the user to be recommended and / or the current space-time feature, and the semantics of the target query word is associated with the semantics of the original query word and the semantics of the auxiliary feature. Finally, the target query word is recommended to the user to be recommended. In this way, the original recommendation word is automatically rewritten according to the auxiliary feature, so that the semantic information conveyed by the finally generated target query word is more rich, and the target query word can be more attractive to a certain extent, thereby improving the recommendation effect when the target query word is recommended.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of network, in particular to a query word processing method and device, electronic equipment and readable storage medium. BACKGROUND

[0002] At present, in order to facilitate users to search network objects on network platform, query words are often recommended to users, and users can use the recommended query words as search keywords to search network objects.

[0003] In the prior art, original query words adapted to users are generated based on historical data of users to directly recommend the original query words to users. In this way, the recommendation effect of query words is poor. SUMMARY

[0004] The present application provides a query word processing method, device, electronic equipment and readable storage medium to solve the problem of poor recommendation effect of query words.

[0005] In a first aspect, the present application provides a query word processing method, which comprises:

[0006] obtaining original query words adapted to a user to be recommended;

[0007] adjusting the original query words according to auxiliary features to generate target query words; the auxiliary features include object features of network objects hit by the original query words, user attribute features of the user to be recommended and / or current space-time features; the semantics of the target query words are associated with the semantics of the original query words and the semantics of the auxiliary features;

[0008] recommending the target query words to the user to be recommended.

[0009] In a second aspect, the present application provides a query word processing device, which comprises:

[0010] a first obtaining module for obtaining original query words adapted to a user to be recommended;

[0011] a generating module for adjusting the original query words according to auxiliary features to generate target query words; the auxiliary features include object features of network objects hit by the original query words, user attribute features of the user to be recommended and / or current space-time features; the semantics of the target query words are associated with the semantics of the original query words and the semantics of the auxiliary features;

[0012] a recommending module for recommending the target query words to the user to be recommended.

[0013] In a third aspect, the present application provides an electronic device, comprising a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the processor implements the method described above when executing the program.

[0014] In a fourth aspect, the present application provides a readable storage medium, which enables an electronic device to execute the method described above when instructions in the storage medium are executed by a processor of the electronic device.

[0015] In the embodiments of the present application, the original query word adapted to the user to be recommended is obtained, and the original query word is adjusted according to the auxiliary feature to generate a target query word. The auxiliary feature includes the object feature of the network object hit by the original query word, the user attribute feature of the user to be recommended, and / or the current space-time feature. The semantics of the target query word is associated with the semantics of the original query word and the semantics of the auxiliary feature. Finally, the target query word is recommended to the user to be recommended. In this way, the original query word is further rewritten according to the auxiliary feature, so that the semantic information conveyed by the finally generated target query word is more rich, and the target query word can be more attractive to a certain extent, thereby improving the recommendation effect when the target query word is recommended. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 is a step flow chart of a query word processing method provided by an embodiment of the present application;

[0018] Figure 2 is a processing flow schematic diagram provided by an embodiment of the present application;

[0019] Figure 3 is another processing flow schematic diagram provided by an embodiment of the present application;

[0020] Figure 4 is a structural diagram of a query word processing device provided by an embodiment of the present application;

[0021] Figure 5 is a structural diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0022] 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, not all, of the embodiments of the present invention. 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.

[0023] Figure 1 This is a flowchart of the steps of a query term processing method provided in an embodiment of the present invention, as follows: Figure 1 As shown, the method may include:

[0024] Step 101: Obtain the original query terms that are suitable for the users to be recommended.

[0025] In this embodiment of the invention, the user to be recommended can be one or more users. For example, the user to be recommended can be some or all users on the network platform. For any user to be recommended, the original query term can be determined based on the user's historical behavior data. For example, based on the user's historical behavior data, the historical search terms searched by the user to be recommended, the trending search terms that the user is interested in, and / or the topic feature terms that the user is interested in can be recalled and added to the candidate query term set. Of course, the trending search terms of the user's location on the network platform can also be directly added to the candidate query term set. Among them, the trending search terms can be words that have been searched more than a preset threshold on the network platform, and the topic feature terms can be used to characterize the topics of network objects that the user to be recommended is interested in. For example, if the user to be recommended frequently orders light-flavored food, then the topic feature term can be "light-flavored"; if the user to be recommended frequently orders items with discounts, then the topic feature term can be "large discount".

[0026] Furthermore, the candidate terms in the query term set can be coarsely sorted. For example, based on parameters such as the search time and click-through rate, terms with more recent search times and higher click-through rates can be ranked higher. Then, the sorted candidate terms can be truncated to obtain the original query terms. For example, the first N candidate terms can be determined as the original query terms, and candidate terms after the Nth term can be removed. In this way, the quality of the original query terms can be ensured to some extent, while avoiding the problem of excessive workload in subsequent steps due to too many original query terms.

[0027] Step 102, adjusting the original query word according to an auxiliary feature to generate a target query word; the auxiliary feature includes an object feature of a network object hit by the original query word, a user attribute feature of the user to be recommended, and / or a current space-time feature; the semantics of the target query word are associated with the semantics of the original query word and the semantics of the auxiliary feature.

[0028] In the embodiment of the application, the specific type of the auxiliary feature can be set according to actual needs. For the object feature of the network object hit by the original query word in the auxiliary feature, since the network object hit by the original query word can be regarded as a supply provided by the original query word to the user, the object feature can also be regarded as a supply feature of the original query word. Further, the object feature of the hit network object can represent the characteristics of these network objects. Therefore, adjusting the original query word based on the object feature can enable the adjusted target query word to convey the basic semantics of the original query word while further conveying the semantics that can reflect the object feature, so that the user to be recommended can conveniently obtain the characteristics of the network objects that can be searched by the target query word based on the target query word, thereby improving the attraction to the user to be recommended, facilitating the selection of the user to be recommended, improving the click rate of the target recommendation word, and prolonging the browsing time of the user.

[0029] Further, for the user attribute feature of the user to be recommended and the current space-time feature in the auxiliary feature, since the user attribute feature can represent the personal attributes of the user to be recommended, and the current space-time feature can represent the characteristics of the time and the space position where the user to be recommended is currently located. Therefore, adjusting the original query word based on the user attribute feature can enable the adjusted target query word to convey the basic semantics of the original query word while further conveying the semantics that conform to the attributes of the user to be recommended or further conveying the semantics that fit the current external environment, thereby improving the attraction to the user to be recommended to a certain extent, facilitating the selection of the user to be recommended, improving the click rate of the target recommendation word, and prolonging the browsing time of the user.

[0030] Step 103, recommending the target query word to the user to be recommended.

[0031] In the embodiment of the application, the target query word can be displayed to the user to be recommended in a preset interface to achieve recommendation. For example, the target query word can be displayed in a history search bar, a search discovery bar, a store entry word bar, or a ranking list bar in a network platform interface. By recommending the target query word that the user to be recommended may be interested in to the user to be recommended, the user to be recommended can further click the target query word to display the network objects hit by the target query word in a landing page, that is, the landing page entered by clicking the target query word and the target query word are associated with each other.

[0032] The query word processing method provided by the embodiment of the present application acquires an original query word adapted to a user to be recommended, adjusts the original query word according to auxiliary features to generate a target query word. The auxiliary features include object features of network objects hit by the original query word, user attribute features of the user to be recommended and / or current space-time features, and the semantics of the target query word is associated with the semantics of the original query word and the semantics of the auxiliary features. Finally, the target query word is recommended to the user to be recommended. In this way, the original query word is automatically rewritten according to the auxiliary features, so that the semantic information conveyed by the finally generated target query word is more rich, and the target query word can be more attractive to a certain extent, thereby improving the recommendation effect when the target query word is used for recommendation.

[0033] Further, compared with the rewriting manner by manual configuration, the original query word is automatically rewritten in the embodiment of the present application, which can save labor cost to a certain extent and improve rewriting efficiency. At the same time, the problem that the rewriting effect is poor due to the influence of personal subjective views in manual rewriting can be avoided.

[0034] Optionally, after the original query word adapted to the user to be recommended is acquired, before the original query word is adjusted according to the auxiliary features to generate the target query word, the following steps can be further executed:

[0035] Step S21, for any original query word, the target number of network objects hit by the original query word is determined.

[0036] In this step, the original query word can be matched with keywords of each network object in a preset index library, and then the matched network object is determined as the network object hit by the original query word. Then, the target number can be determined based on the number of hit network objects. The preset index library can include a POI index and an SPU index. In an implementation manner, the POI can point to a store in a network platform, and the SPU can point to an item in the store in the network platform. The store and the item in the store can both be the network object mentioned above. Further, the inverted index and the forward index can be established based on the keywords of the POI in advance, and the inverted index and the forward index can be established based on the keywords of the SPU to construct the index library.

[0037] Step S22, in the case that the target number is less than a preset number threshold, the original query word is eliminated.

[0038] In this step, the preset quantity threshold can be set according to actual needs, and the embodiment of the application does not limit this. Further, if the target quantity is greater, it can be indicated that the number of network objects that can be searched by searching with the original query word is greater, and the supply quantity of the original query word is greater. On the contrary, if the target quantity is smaller, it can be indicated that the number of network objects that can be searched by searching with the original query word is smaller, and the supply quantity of the original query word is smaller. Further, if the target quantity is smaller than the preset quantity threshold, it can be determined that the supply quantity of the original query word is too small, and therefore the original query word can be deleted. The remaining original query words after elimination can constitute an original query word effective set. Accordingly, subsequent processing can be performed only on the original query words in the original query word effective set.

[0039] Further, the query word in the embodiment of the application can be a one-hop word, and the action of a user clicking the one-hop word can be referred to as two-hop, and accordingly, the page jumped to can be referred to as a two-hop page. The network objects hit by the query word can be limited in the two-hop page. In the embodiment of the application, by eliminating the original query word with a small supply quantity in advance, i.e., eliminating the original query word with a target quantity of hit network objects smaller than the preset quantity threshold, the subsequent recommendation of the query word with a small supply quantity to the user to be recommended can be avoided, which can avoid the problem that the content that can be displayed in the two-hop page is too small, and the demand of the one-hop word cannot be efficiently met, thereby causing the search effect of the query word to be poor. At the same time, by filtering the original query word with a small supply quantity before adjusting the original query word, the workload of subsequent adjustment operations can be reduced to some extent, and the processing resources required subsequently can be saved.

[0040] Optionally, the operation of determining the target quantity of the network objects hit by the original query word can specifically include:

[0041] Step S31, obtaining a preset recommendation period; the preset recommendation period is used to represent a period of recommending the target query word.

[0042] In this step, the preset recommendation period can be set according to actual needs, for example, the preset recommendation time can be set to X hours after the current time. Further, when the preset recommendation period is obtained, the preset period value set in advance can be directly read as the preset recommendation period.

[0043] Step S32, determining the number of target network objects in the network objects hit by the original query word as the target quantity; the target network object is in an online state in the preset recommendation period.

[0044] In this step, the original query word can be matched with the keywords of each network object in the preset index library to determine the matched network object. Then, the network object in the online state in the preset recommendation period in the matched network object is determined as the target network object. Then, the number of target network objects can be counted to obtain the target number.

[0045] The positive index in the preset index library can be constructed based on the online state of each network object, that is, the online period of the network object can be determined based on the positive index, and then the online state of the network object in the preset recommendation period can be conveniently determined. The network object in the online state can mean that the network object is in an available state. For example, the store or the item in the take-out platform is in an orderable state in the business hours, that is, in the online state. Outside the business hours, it is in an unorderable state, that is, in a non-online state.

[0046] In the embodiment of the application, the preset recommendation period is obtained, and the number of network objects in the online state in the preset recommendation period in the hit network object is determined as the target number. In this way, the target number can accurately represent the actual number of network objects that can be searched based on the target query word, that is, the real supply amount can be accurately represented, and the filtered original query word can be more adapted to the actual recommendation demand.

[0047] Optionally, the operation of recommending the target query word to the user to be recommended can include:

[0048] Step S41, for any target query word, determining a click rate score corresponding to the target query word; the click rate score is positively correlated with the first estimated click rate of the target query word.

[0049] In this step, the first estimated click rate can be determined based on a preset Click-Through-Rate (CTR) estimation model. Specifically, the relevant information of the target query word can be input into the CTR estimation model, and then the output of the CTR estimation model can be taken as the first estimated click rate PCTR i . Further, the click rate score of the target query word can be determined based on the first estimated click rate of the target query word.

[0050] Step S42, selecting the first M target query words with the maximum click rate score, and recommending the first M target query words to the user to be recommended.

[0051] In this step, the target query words can be sorted according to the click rate scores corresponding to the target query words, to obtain a sorting result. Then, according to the sorting result, the first M target query words with the maximum click rate scores are selected. The specific value of M can be set according to actual requirements, and the embodiments of the present application do not make any limitation in this regard. For example, the target query words can be sorted in descending order of the click rate scores, and correspondingly, the first M target query words in the sorting result are taken, that is, the first M target query words with the maximum click rate scores are obtained. Further, the target query words can also be sorted in ascending order of the click rate scores, and correspondingly, the last M target query words in the sorting result are taken, that is, the first M target query words with the maximum click rate scores are obtained.

[0052] In the embodiments of the present application, by determining the click rate scores corresponding to the target query words, the click rate scores can be positively correlated with the first estimated click rates of the target query words, the first M target query words with the maximum click rate scores are selected, and the first M target query words are recommended to the user to be recommended, which can to some extent avoid the recommended target query words being too many, leading to the difficulty of the user to be recommended, and improve the click rate of the recommended target query words.

[0053] Optionally, the operation of determining the click rate score corresponding to the target query word can include:

[0054] In step S51, a feature click rate corresponding to the object feature is determined; the feature click rate is positively correlated with the second estimated click rate of the target network object, and the target network object is a network object with the object feature in the hit network object.

[0055] In this step, for any object feature, the related information of the target network object with the object feature can be taken as the input of the CTR estimation model, and then the output of the CTR estimation model is taken as the second estimated click rate of the target network object. Further, the sum of the second estimated click rates of all target network objects can be determined as the feature click rate corresponding to the object feature.

[0056] For example, it is assumed that the target query word query i The hit network objects include poi1, poi2,..., and poi x The object features include C1, C2,..., and C n The target network objects with C1 include [poi1, poi2,..., and poi x The target network objects with C2 include [poi1, poi2,..., and poi x] The target network objects with C n include [poi1, poi2,..., and poi nThen the sum of the second estimated click rates of poi1, poi2, …, and poi x is determined as the feature click rate corresponding to C2, and the sum of the second estimated click rates of poi1, poi2, …, and poi n is determined as the feature click rate corresponding to C3. The object features possessed by the network objects hit by the target query term can also be referred to as provided features. The target query term, the hit network objects, and the possessed object features can be represented in the form of a set as follows:

[0057] query i : {C1: [poi1, poi2, …, and poi n ], C2: [poi1, poi2, …, and poi n ], …, C n : [poi1, poi2, …, and poi n ]

[0058] In step S52, the click rate score is generated according to the first estimated click rate, the feature click rate, and the target quantity of the network objects hit by the original query term. The click rate score is positively correlated with the feature click rate and negatively correlated with the target quantity.

[0059] In this step, the first estimated click rate, the feature click rate, and the target quantity can be input into a preset score calculation formula as parameters, and then the value of the score calculation formula is taken as the click rate score. The score calculation formula can be set according to actual requirements. For example, the score calculation formula can be as follows:

[0060]

[0061] wherein PCTR i represents the first estimated click rate, ctr ij represents the jth

[0062] object feature of the ithtarget query term, POSweight ij represents the preset weight of the jthobject feature of the ithtarget query term, and len i represents the target quantity of the ithtarget query term. Since the target quantities of different target query terms can be different, and the target quantity will affect the click rate to some extent, in the embodiment of the present application, the click rate score is calculated in a manner that the click rate score is negatively correlated with the target quantity, so that the click rate score can more fairly represent the click rate to some extent.

[0063] It should be noted that the exposure data and the click data of the user to the displayed target query word can also be used as the buried point data in the embodiment of the application. After the target query word is displayed, the display data and the buried point data of the target query word are obtained. Then, the display data and the buried point data are returned to the CTR estimation model to optimize the CTR estimation model, thereby improving the accuracy of the predicted click rate.

[0064] In the embodiment of the application, the feature click rate corresponding to the original query word is determined, the feature click rate is the click rate corresponding to the object feature, and the click rate corresponding to the object feature is positively correlated with the second estimated click rate of the network object with the object feature. Then, the click rate score is generated according to the first estimated click rate, the feature click rate, and the target number of the network object hit by the original query word; the click rate score is positively correlated with the feature click rate and negatively correlated with the target number. In this way, the click rate score is calculated by further combining the feature click rate and the target number, which can to some extent enable the click rate score to more accurately measure the click rate that the target query word itself can bring, thereby improving the accuracy of the subsequent selection operation.

[0065] Optionally, the auxiliary feature can also be obtained by the following steps before the original query word is adjusted according to the auxiliary feature to generate the target query word in the embodiment of the application.

[0066] In step S61, the object feature label, the user attribute label of the user to be recommended, the current time information and / or the current location information are obtained; the object feature label is used to represent the object feature; the user attribute label is used to represent the user attribute feature; and the current time information and the current location information are used to represent the space-time feature.

[0067] In this step, the specific types of object features can be set according to actual needs. The object features can include features in multiple dimensions, and different object features can be represented by different object feature labels. For example, the object feature labels can include "brand", "promotion", "high repeat purchase", "taste", etc. The object features possessed by each network object can be determined by the characteristics of the network object itself, and the object feature labels possessed by different network objects can be different. Further, the user attribute label of the user to be recommended can be set in advance according to the personal information of the user. Accordingly, the user attribute label of the user to be recommended can be found from a preset data management platform (DMP) based on the user identifier of the user to be recommended. For example, the user attribute label can include "young people", "white-collar workers", "bourgeois youth", etc.

[0068] Further, the current time information and / or the current location information can be acquired in real time. The current time information can be a specific time point, or information associated with the current time. For example, the current time information can include meal time (e.g., breakfast, lunch, dinner, supper, afternoon tea), season (e.g., spring, summer, autumn, winter), solar term (e.g., winter solstice, Gu Yu, autumnal equinox, etc.), festival (Qixi, Spring Festival, Valentine's Day, etc.). Further, the current location information can be a specific location, or information associated with the specific location. For example, the current location information can include a commercial district, a street, a city, or the like. The current time information and the current location information can be acquired through the current environment information.

[0069] In the embodiments of the present application, the object features are represented by the object feature labels, the user attribute features are represented by the user attribute labels, and the space-time features are represented by the current time information and the current location information. By acquiring the object feature labels, the user attribute labels of the to-be-recommended user, the current time information and / or the current location information, the auxiliary features can be conveniently acquired, and the acquisition cost can be reduced and the acquisition efficiency can be improved to a certain extent.

[0070] Meanwhile, the user attribute labels of the to-be-recommended user, the current time information and the current location information are associated with the to-be-recommended user, and therefore, the subsequent user attribute labels, the current time information and the current location information as auxiliary features can make the target query words rewritten based on the auxiliary features meet the user preferences to a certain extent, and thus ensure the rewriting effect.

[0071] Optionally, before acquiring the object feature labels, the embodiments of the present application can further perform the following operations:

[0072] In step S71, whether the network objects have the object features of each specified dimension is detected according to the related information of each network object.

[0073] In this step, the specific types of related information and the specified dimensions can be set according to actual needs. For example, the related information can include name, participated activities, sales, repeat purchase rate, discount amount, discount method, brand qualification, and the like. Further, whether the network object satisfies the required rule of the object feature of the specified dimension can be determined according to the related information, so as to ensure whether the object feature of the specified dimension is possessed. For example, for the object feature "high repeat purchase" of the specified dimension, the monthly sales and the repeat purchase rate of the network object can be obtained from the related information. If the monthly sales of the network object > x and the repeat purchase rate > y, it can be determined that the network object satisfies the rule and possesses the object feature "high repeat purchase". For the object feature "promotion" of the specified dimension, whether the network object has a discount activity can be determined based on the related information. If yes, it can be determined that the network object satisfies the rule and possesses the object feature "promotion". For the object feature "brand" of the specified dimension, whether the network object has brand qualification can be determined based on the related information. If yes, it can be determined that the network object satisfies the rule and possesses the object feature "brand". For the object feature "Sichuan cuisine" of the specified dimension, the number of Sichuan cuisine in the network object can be obtained from the related information. If the number satisfies a preset rule, it can be determined that the network object satisfies the rule and possesses the object feature "Sichuan cuisine".

[0074] In step S72, for any specified dimension, if the network object possesses the object feature of the specified dimension, the object feature label of the specified dimension is set for the network object to generate a feature database.

[0075] For example, for the network object possessing the object feature "high repeat purchase", the "high repeat purchase" label can be set. For the network object possessing the object feature "brand", the "brand" label can be set. For the network object possessing the object feature "promotion", the "promotion" label can be set. Further, the feature vector can be generated based on the object feature label possessed by the network object, so as to quickly determine the object feature possessed by the network object. Each element of the feature vector can correspond to an object feature of a specified dimension. When the network object possesses the object feature of the specified dimension corresponding to the element, the value of the element can be set as a first preset value. On the contrary, when the network object does not possess the object feature of the specified dimension corresponding to the element, the value of the element can be set as a second preset value. For example, the first preset value can be 1 and the second preset value can be 0. For the feature vector (1, 0, 1, 0, 0), it can represent that the network object possesses the object feature of the specified dimension corresponding to the first element and the third element. Further, in the embodiment of the present application, the related information can also be directly used as the input of the preset model, and the feature label of the network object is mapped based on the preset model.

[0076] Correspondingly, the operation of obtaining the object feature label can specifically include: searching, from the feature database, for the object feature label corresponding to the network object hit by the original query word. Specifically, the feature vector of the network object hit by the original query word can be searched from the feature database according to the identifier of the network object hit by the original query word. For example, the identifier of the network object hit by the original query word is input into a preset query algorithm to search from the feature database. Then, the label of the object feature corresponding to the element with the first preset value in the feature vector is obtained, and the object feature label corresponding to the network object hit by the original query word is obtained.

[0077] In the embodiment of the application, whether the network object has the object feature of each specified dimension is detected according to the related information of each network object in advance. Then, for any specified dimension, the object feature label of the specified dimension is set for the network object in the case that the network object has the object feature of the specified dimension, so as to generate the feature database. Correspondingly, when the object feature label is obtained, the object feature label corresponding to the network object hit by the original query word is only searched from the feature database, so that the obtaining operation can be realized, and the convenience of obtaining the object feature label can be ensured to a certain extent, thereby improving the processing efficiency.

[0078] Optionally, the operation of adjusting the original query word according to the auxiliary feature to generate the target query word in the embodiment of the application can specifically include:

[0079] Step S81: in the case that the original query word is used to represent a network object, determining the type to which the network object belongs to obtain a target type.

[0080] In this step, if the original query word is used to represent a network object, the original query word can be determined as a normal query word associated with a user-related attribute. Further, the network object represented by the original query word can be an object pointed by the semantics of the original query word. For example, the semantics object represented by the original query word can be an entity, such as dumplings, milk tea, mobile phones, computers, and the like. Further, the type of the semantics object can be preset. For example, the type of the semantics object can include staple food, beverage, electronic product, and the like. Specifically, the type to which the network object represented by the original query word belongs can be determined according to a preset type-entity correspondence relationship, and then the target type is obtained.

[0081] Step S82: determining a target auxiliary feature corresponding to the original query word according to a preset correspondence relationship between the type and the auxiliary feature and the target type; the target auxiliary feature is an auxiliary feature with a collocation degree satisfying a preset condition with the original query word.

[0082] In this step, the correspondence between the preset type and the auxiliary feature can be implemented through a preset rule base, in which the correspondence between the type and the auxiliary feature, i.e., the combination relationship of the query word and the feature, can be defined. When the correspondence is preset, the auxiliary feature that meets the preset condition in terms of the degree of collocation with the network object of the type can be set for the type. The preset condition can be set according to actual needs. Further, the auxiliary feature corresponding to the target type can be found in the correspondence to serve as the target auxiliary feature. For example, assuming that the space-time feature corresponding to the "main course" type is "solar term", the space-time feature corresponding to the "drink" type is "season", and the user attribute feature. Then, in the case where the original query word represents the network object as "dumpling", the space-time feature "solar term" is determined as the target auxiliary feature. In the case where the original query word represents the network object as "milk tea" and "coffee", the space-time feature "season" and the user attribute feature are determined as the target auxiliary feature.

[0083] In step S83, the original query word is rewritten according to the semantics represented by the target auxiliary feature to obtain the target query word.

[0084] Since the degree of collocation between the target auxiliary feature and the original query word meets the preset condition, i.e., the degree of collocation is high, the way of rewriting the original query word based on the semantics represented by the target auxiliary feature can ensure the quality of the target query word obtained by rewriting to some extent.

[0085] In the embodiment of the application, in the case where the original query word is used to represent the network object, the type to which the network object belongs is determined to obtain the target type; the target auxiliary feature corresponding to the original query word is determined according to the correspondence between the preset type and the auxiliary feature and the target type; the target auxiliary feature is an auxiliary feature that meets the preset condition in terms of the degree of collocation with the original query word; and the original query word is rewritten according to the semantics represented by the target auxiliary feature to obtain the target query word. In this way, by predefining the correspondence between the type and the auxiliary feature, the target auxiliary feature that has a better collocation effect is matched for the original query word, and the original query word is rewritten based on the target auxiliary feature, thereby ensuring the query word rewriting effect to some extent.

[0086] Optionally, the operation of rewriting the original query word according to the semantics represented by the target auxiliary feature to obtain the target query word can specifically include: combining the target auxiliary feature and the original query word to obtain the target query word. In this way, rewriting can be achieved through combination, and rewriting efficiency can be ensured to some extent. Specifically, the target auxiliary feature and the original query word can be taken as inputs of a preset rewriting algorithm, the preset rewriting algorithm can combine the target auxiliary feature and the original query word through a preset conjunction word to ensure the fluency of the target query word. Alternatively, the content of the target auxiliary feature can be adjusted, and the adjusted content and the original query word can be combined through a preset conjunction word. The adjusted content and the target auxiliary feature are semantically associated. Finally, the output of the preset rewriting algorithm can be taken as the target query word. For example, assuming that the current solar term is "winter solstice", the original query word "jiaozi" and "winter solstice" can be combined to obtain the target query word "winter solstice eats jiaozi". Assuming that the current season is "autumn", the original query word "milk tea" and "autumn" can be combined to obtain the target query word "the first cup of milk tea in autumn". Assuming that the user attribute label is "white-collar", the original query word "coffee" and "white-collar" can be combined to obtain the target query word "coffee that white-collar people love to drink". Further, the original query word "milk tea" and the object feature "quality" can be combined to obtain the target query word "quality milk tea", or the object feature "high repeat purchase" can be combined to obtain the target query word "milk tea you often drink". The original query word "coffee" and the time-space feature "afternoon" can be combined to obtain the target query word "a cup of coffee after a nap".

[0087] Optionally, the operation of adjusting the original query word according to the auxiliary feature to generate the target query word can further include the following steps:

[0088] Step S91: In the case where the original query word is used to represent a theme possessed by the network object, an object feature matching the theme is obtained.

[0089] Step S92: The original query word is rewritten according to the feature value of the matching object feature to obtain the target query word.

[0090] In the embodiment of the present application, if the original query word is used to represent the theme possessed by the network object, it can be determined that the original query word is the subject query word derived from the user-related attribute. Further, the object feature that is adapted to the theme, i.e., the object feature with similar represented semantics, can be obtained from the object features possessed by the network object hit by the original query word. For example, in the case of the theme "large amount of discount", the object feature "discount amount" that matches the theme "large amount of discount" can be obtained. In the case of the theme "large amount of discount", the object feature "discount" that matches the theme "large amount of discount" can be obtained. Further, the feature value of the matching object feature can be obtained from the related information of the network object hit by the original query word. For example, the specific value of the discount amount of the hit network object and the specific value of the discount can be obtained. Further, the original query word can be replaced by the feature value of the object feature. For example, "large amount of discount" can be replaced by "discount 50 by 20", and "large amount of discount" can be replaced by "minimum 1 discount". The original query word "Sichuan cuisine" is replaced by the specific value "no spicy food" of the matching object feature "taste". It should be noted that since the hit network object is often multiple, the specific value of the feature value of the object feature can be multiple, and accordingly, a specific value can be randomly selected or a specific value with the highest frequency of occurrence or a specific value with the smallest value can be selected for replacement.

[0091] In the embodiment of the present application, in the case where the original query word is used to represent the theme possessed by the network object, the object feature that matches the theme is obtained; and the original query word is rewritten based on the feature value of the matching object feature to obtain the target query word. Since the object feature that matches the theme can more finely and accurately represent the theme, the original query word is rewritten based on the feature value of the matching object feature, which can ensure the rewriting effect to some extent.

[0092] It should be noted that the embodiment of the present application can also filter the target query word based on the number of network objects hit by each target query word after obtaining the target query word, to obtain the final target query word effective set, so as to further ensure that the target query word displayed to the user has sufficient supply. The implementation manner of filtering the target query word based on the number of network objects hit by each target query word can refer to the implementation manner of filtering the original query word based on the number of network objects hit by each original query word described above. For example, Figure 2 is a processing flow diagram provided by the embodiment of the present application, as Figure 2As shown, the query term retrieval module first obtains the original query terms. Then, the supply filtering module filters the original query terms according to a preset index. Next, the corpus rewriting module rewrites the filtered original query terms to obtain the target query terms. Finally, the prediction ranking module ranks the target query terms according to their corresponding click-through rate (CTR) scores. The "front-end display" stage then shows the top M target query terms with the highest CTR scores.

[0093] Furthermore, Figure 3 This is a schematic diagram of another processing flow provided by an embodiment of the present invention, such as... Figure 3 As shown, for the original query term "milk tea" that matches the users to be recommended, it can be rewritten as the target query term within the dashed box. Next, through sorting and output, the target query term "high-quality brand milk tea shops" is finally displayed to the users to be recommended. Finally, based on the click actions of the users to be recommended, event tracking data can be obtained to optimize the CTR prediction model.

[0094] Figure 4 This is a structural diagram of a query term processing device provided in an embodiment of the present invention. The device 20 may include:

[0095] The first acquisition module 201 is used to acquire the original query terms that are suitable for the users to be recommended;

[0096] The generation module 202 is used to adjust the original query term according to auxiliary features to generate a target query term; the auxiliary features include the object features of the network object hit by the original query term, the user attribute features of the user to be recommended, and / or the current spatiotemporal features; the semantics of the target query term is related to the semantics of the original query term and the semantics of the auxiliary features;

[0097] The recommendation module 203 is used to recommend the target query term to the user to be recommended.

[0098] Optionally, the device 20 further includes:

[0099] The determination module is used to determine the number of target network objects hit by any of the original query terms;

[0100] The elimination module is used to eliminate the original query terms when the number of targets is less than a preset threshold.

[0101] Optionally, the determining module is specifically used for:

[0102] Obtain a preset recommendation time period; the preset recommendation time period is used to represent the time period during which the target query term is recommended;

[0103] determine a number of target network objects hit by the original query word in the network objects as the target number; the target network objects are in an online state in the preset recommendation period.

[0104] Optionally, the recommendation module 203 is specifically used for:

[0105] For any target query word, determine a click rate score corresponding to the target query word; the click rate score is positively correlated with a first estimated click rate of the target query word;

[0106] select the first M target query words with the maximum click rate scores, and recommend the first M target query words to the user to be recommended.

[0107] Optionally, the recommendation module 203 is further specifically used for:

[0108] determine a feature click rate corresponding to the object feature; the feature click rate is positively correlated with a second estimated click rate of a target network object; the target network object is a network object with the object feature in the hit network objects;

[0109] generate the click rate score according to the first estimated click rate, the feature click rate, and a target number of the network objects hit by the original query word; the click rate score is positively correlated with the feature click rate and negatively correlated with the target number.

[0110] Optionally, the generation module 202 is specifically used for:

[0111] in a case where the original query word is used to represent a network object, determine a type to which the network object belongs, to obtain a target type;

[0112] determine a target auxiliary feature corresponding to the original query word according to a preset correspondence between a type and an auxiliary feature and the target type; the target auxiliary feature is an auxiliary feature with a preset condition met in terms of a collocation degree with the original query word;

[0113] rewrite the original query word according to a semantic represented by the target auxiliary feature, to obtain the target query word.

[0114] Optionally, the generation module 202 is further specifically used for:

[0115] combine the target auxiliary feature with the original query word, to obtain the target query word.

[0116] Optionally, the generation module 202 is further specifically used for:

[0117] In the case that the original query word is used to represent a theme possessed by the network object, an object feature matching the theme is acquired;

[0118] The original query word is rewritten according to a feature value of the matching object feature, to acquire the target query word.

[0119] Optionally, the apparatus 20 further comprises:

[0120] The second acquisition module is configured to acquire the object feature label, a user attribute label of the user to be recommended, current time information and / or current location information; the object feature label is used to represent the object feature; the user attribute label is used to represent the user attribute feature; and the current time information and the current location information are used to represent the spatio-temporal feature.

[0121] Optionally, the apparatus 20 further comprises:

[0122] The detection module is configured to detect, according to the related information of each network object, whether the network object possesses an object feature of each specified dimension;

[0123] The setting module is configured to, for any specified dimension, if the network object possesses the object feature of the specified dimension, set an object feature label of the specified dimension for the network object, to generate a feature database.

[0124] The second acquisition module is specifically configured to: find, from the feature database, an object feature label corresponding to a network object hit by the original query word.

[0125] The query word processing apparatus provided by the embodiment of the application acquires an original query word adapted to a user to be recommended, adjusts the original query word according to auxiliary features, to generate a target query word. The auxiliary features include an object feature of a network object hit by the original query word, a user attribute feature of the user to be recommended and / or a current spatio-temporal feature. The semantics of the target query word is associated with the semantics of the original query word and the semantics of the auxiliary features. Finally, the target query word is recommended to the user to be recommended. In this way, the original query word is automatically rewritten according to the auxiliary features, so that the semantic information conveyed by the finally generated target query word is more abundant, and the target query word is more attractive to a certain extent, thereby improving the recommendation effect when the target query word is used for recommendation.

[0126] Further, compared with the rewriting manner by manual configuration, the original query word is automatically rewritten in the embodiment of the application, which can save labor cost and improve rewriting efficiency to a certain extent. Meanwhile, the problem that the rewriting effect is poor due to the influence of personal subjective views in manual rewriting can be avoided.

[0127] The application also provides an electronic device, referring to Figure 5 comprising a processor 301, a memory 302, and a computer program 3021 stored in the memory and capable of running on the processor, and the processor implements the method of the foregoing embodiments when running the program.

[0128] The application also provides a readable storage medium, when the instructions in the storage medium are executed by the processor of an electronic device, the electronic device can execute the method of the foregoing embodiments.

[0129] For the device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts are referred to the part of the method embodiments.

[0130] The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other apparatus. Various general purpose systems can be used with these teachings, with the structure for a variety of such systems will be apparent from the description above. In addition, the present application is not described with reference to any particular programming language. It will be appreciated that a variety of programming languages can be used to implement the teachings of the application as described herein, and any references below to specific languages are provided for disclosure of enablement only.

[0131] In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the application can be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been described in detail in order not to obscure the understanding of this description.

[0132] Similarly, it is to be understood that the brunt of the technical details of the application are often split up into separate embodiments, figures, or descriptions of these for the purpose of brevity and to help understand one or more of the various inventive aspects. However, this method of disclosure is not to be interpreted as reflecting an intention that the application requires more features than are explicitly recited in each claim. Rather, inventive aspects lie in less than all features of the single disclosed embodiments. Accordingly, the claims, as follows, reflect applicant's consideration of possibilities of this kind. Thus, the claims are hereby expressly incorporated into this detailed description of the specific embodiments of the application, with each claim standing on its own as a separate embodiment of this application.

[0133] Those skilled in the art will appreciate that the modules in the apparatuses in the embodiments can be adapted and placed in one or more apparatuses other than that of the embodiments. The modules or units or components in the embodiments can be combined into one module or unit or component, and furthermore can be split into multiple sub-modules or sub-units or sub-components. Any combination of all the features disclosed in the specification (including the accompanying claims, abstract and drawings), and any method or process or device of the embodiments disclosed can be taken, except that at least some of such features and / or processes or units are mutually exclusive, unless explicitly stated otherwise. Each feature disclosed in the specification (including the accompanying claims, abstract and drawings), can be replaced by alternative features providing the same, equivalent or similar functions unless explicitly stated otherwise.

[0134] Embodiments of the various components of the application can be implemented in hardware, or as software modules running in one or more processors, or combinations thereof. Those skilled in the art will appreciate that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all of the functionality of some or all of the components in the sequencing apparatus according to the application. The application can also be implemented as a program for executing part or all of the methods described herein on a device or apparatus. Such program(s) can be stored on computer readable media or can be transmitted over a network, for example, from an Internet site, or can be provided on a carrier signal.

[0135] It should be noted that the above-mentioned embodiments illustrate rather than limit the application, and that those skilled in the art will be able to design many alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word 'comprising' does not exclude the presence of elements or steps other than those listed in a claim. The word 'a' or 'an' preceding an element does not exclude the presence of a plurality of such elements. The application can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the system claims enumerating several means, several of these means can be embodied by one and the same item of hardware. The use of the words 'first','second', and 'third', etc. do not imply any order. These words are to be understood as names.

[0136] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0137] The above merely describes preferred embodiments of the present application and is not used to limit the present application, and any modification, equivalent replacement, and improvement within the spirit and principle of the present application should be included in the protection scope of the present application.

[0138] The above merely describes preferred embodiments of the present application and is not used to limit the present application, and any modification, equivalent replacement, and improvement within the spirit and principle of the present application should be included in the protection scope of the present application.

[0138] The above merely describes preferred embodiments of the present application and is not used to limit the present application, and any modification, equivalent replacement, and improvement within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A query word processing method characterized by, The method comprises: acquiring an original query word adapted to a user to be recommended; adjusting the original query word according to auxiliary features to generate a target query word; the auxiliary features comprise object features of network objects hit by the original query word, user attribute features of the user to be recommended, and / or current space-time features; the semantics of the target query word are associated with the semantics of the original query word and the semantics of the auxiliary features; for any target query word, determining a feature click rate corresponding to the object features; the feature click rate is positively correlated with a second estimated click rate of a target network object, which is a network object having the object features among the hit network objects; generating a click rate score according to a first estimated click rate of the target query word, the feature click rate, and a target number of network objects hit by the original query word; the click rate score is positively correlated with the first estimated click rate and the feature click rate, and is negatively correlated with the target number; selecting the first M target query words with the largest click rate scores, and recommending the first M target query words to the user to be recommended.

2. The method of claim 1, wherein, After acquiring the original query word adapted to the user to be recommended, and before adjusting the original query word according to the auxiliary features to generate the target query word, the method further comprises: for any original query word, determining a target number of network objects hit by the original query word; in the case where the target number is less than a preset number threshold, eliminating the original query word.

3. The method of claim 2, wherein, The determination of the target number of network objects hit by the original query word comprises: acquiring a preset recommendation period; the preset recommendation period is used to represent a period of recommending the target query word; determining the number of target network objects among the network objects hit by the original query word as the target number; the target network objects are in an online state in the preset recommendation period.

4. The method of claim 1, wherein, The adjustment of the original query word according to the auxiliary features to generate the target query word comprises: in the case where the original query word is used to represent a network object, determining a target type to which the network object belongs; determining a target auxiliary feature corresponding to the original query word according to a preset correspondence between types and auxiliary features and the target type; the target auxiliary feature is an auxiliary feature whose collocation degree with the original query word meets a preset condition; rewriting the original query word according to the semantics represented by the target auxiliary feature to acquire the target query word.

5. The method of claim 4, wherein, The rewriting of the original query word according to the semantics represented by the target auxiliary feature to acquire the target query word comprises: combining the target auxiliary feature with the original query word to obtain the target query word.

6. The method of claim 4, wherein, The method further comprises: in the case where the original query word is used to represent a theme possessed by a network object, acquiring an object feature matched with the theme; rewriting the original query word according to a feature value of the matched object feature to acquire the target query word.

7. The method of claim 1, wherein, The method further comprises: Obtaining an object feature label, a user attribute label of the user to be recommended, current time information and / or current location information; the object feature label is used to represent the object feature; the user attribute label is used to represent the user attribute feature; the current time information and the current location information are used to represent the spatio-temporal feature.

8. The method of claim 7, wherein, Before the object feature label is obtained, the method further comprises: According to the related information of each network object, detecting whether the network object has the object feature of each specified dimension; For any specified dimension, if the network object has the object feature of the specified dimension, setting the object feature label of the specified dimension for the network object to generate a feature database; The object feature label includes: searching the object feature label corresponding to the network object hit by the original query term from the feature database.

9. A query word processing apparatus characterized by comprising: The device comprises: A first obtaining module is configured to obtain an original query term adapted to a user to be recommended; A generating module is configured to adjust the original query term according to auxiliary features to generate a target query term; the auxiliary features include an object feature of a network object hit by the original query term, a user attribute feature of the user to be recommended and / or a current spatio-temporal feature; the semantics of the target query term are associated with the semantics of the original query term and the semantics of the auxiliary features; A recommending module is configured to determine, for any target query term, a feature click rate corresponding to the object feature; the feature click rate is positively correlated with a second estimated click rate of a target network object, the target network object being a network object having the object feature among the network objects hit by the original query term; generate a click rate score according to a first estimated click rate of the target query term, the feature click rate and a target number of network objects hit by the original query term; the click rate score is positively correlated with the first estimated click rate and the feature click rate, and negatively correlated with the target number; select the first M target query terms with the largest click rate scores, and recommend the first M target query terms to the user to be recommended.

10. An electronic device, comprising: Comprise: A processor, a memory and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method of one or more of claims 1-8 when executing the program.

11. A readable storage medium, characterized by, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device can perform the method of one or more of claims 1-8.

Images