Translation optimization method, device and processor

Abstract

The present invention discloses a translation optimization method, device and processor. The method comprises: obtaining a plurality of target query statements and translation results of a plurality of target query statements, wherein the target query statement is a query statement that satisfies a preset condition, and the preset condition is used to characterize the translation accuracy of the query statement in the target domain; training a target machine translation system based at least on the translation results of the plurality of target query statements and the plurality of target query statements; translating the query statements to be translated in the target domain based on the target machine translation system. The present invention solves the technical problem in the related art that when performing cross-language queries in a specific domain, the translation accuracy of the query statement is low, resulting in poor adoption of the query results.

Description

Technical Field

[0001] The present invention relates to the field of translation technology, and in particular to a translation optimization method, device and processor. Background Art

[0002] When entering a search query and searching based on a cross-language search platform, most fields currently lack parallel bilingual query data. It is necessary to first translate the search query into vocabulary in the language of the search platform through a machine translation model, and then query based on the translated vocabulary to obtain the query results. Since the search query is very short and its structure and word order are different from traditional normal sentences, the machine translation model has poor translation results in the absence of translation field data, resulting in a large difference between the search query results and the user's expected results.

[0003] In order to solve the problem of domain adaptation of query statements during search queries, the related art uses bilingual query data in a small number of fields to adjust the parameters of the general statistical machine translation system, or re-ranks the generated best translation candidate words based on the search index to achieve domain adaptation of the query statement. However, the above method does not use the feedback data of the user in the actual application process, and the retrieval quality obtained is poor. In addition, other methods for solving the problem of inaccurate translation query results have also appeared in the related art, for example, using a bilingual dictionary to find web page seeds, and mining the uniform resource locator pair (URL pair) pattern from the click data, and then using the mined URL pair pattern to generate candidate query pairs. Alternatively, use a bilingual dictionary to translate the query statement in the monolingual click data of the target language, and build a translation model of the query statement based on the generated pseudo corpus. Alternatively, in the query expansion task, the input query in one language is mapped to the query statement in another language to use the co-occurrence statistics of bilingual words for translation and query. Alternatively, the click data in the search engine query log is used, combined with the link log clicked by the user to train the ranking algorithm to obtain the best query statement. Alternatively, the title translation quality can be improved based on the user behavior recorded in the implicit task feedback collected from the cross-language search task. Although the above methods use user behavior data in different cross-language search tasks, the database based on which the cross-language query statements are obtained is a conventional database, and the translation results are different from the actual needs of users in applications.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present invention provide a translation optimization method, device and processor to at least solve the technical problem in the related art that when performing cross-language queries in a specific field, the translation accuracy of query statements is low, resulting in poor adoption of query results.

[0006] According to one aspect of an embodiment of the present invention, a translation optimization method is provided, comprising: obtaining a plurality of target query statements and translation results of a plurality of target query statements, wherein the target query statements are query statements that meet preset conditions, and the preset conditions are used to characterize the translation accuracy of the query statements in a target domain; training a target machine translation system based at least on the plurality of target query statements and the translation results of the plurality of target query statements; and translating the query statements to be translated in the target domain based on the target machine translation system.

[0007] According to another aspect of an embodiment of the present invention, a translation optimization device is also provided, including: a first acquisition unit, used to acquire multiple target query statements and translation results of multiple target query statements, wherein the target query statements are query statements that meet preset conditions, and the preset conditions are used to characterize the translation accuracy of the query statements in the target domain; a training unit, used to train a target machine translation system based on at least multiple target query statements and translation results of multiple target query statements; and a translation unit, used to translate the query statements to be translated in the target domain based on the target machine translation system.

[0008] In an embodiment of the present invention, a method of training a machine translation system by using feedback data of query sentences that meet the translation accuracy in a target field to a machine translation system is adopted. By obtaining a target query sentence and its translation result, and using the target query sentence and its translation result as training data to train a target translation system, and then translating the query sentence through the target system, the purpose of improving the translation accuracy of the query sentence when performing cross-language queries in a specific field is achieved, thereby achieving the technical effect of improving the adoption rate of the query results, and further solving the technical problem in the related art that when performing cross-language queries in a specific field, the low translation accuracy of the query sentence leads to poor adoption of the query results. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0010] Figure 1 is a hardware structure block diagram of a computer terminal according to a translation optimization method of an embodiment of the present invention;

[0011] Figure 2 is a flow chart of a translation optimization method according to an embodiment of the present invention;

[0012] Figure 3 is a schematic diagram of another translation optimization method according to an embodiment of the present invention;

[0013] Figure 4 is a schematic diagram of a translation evaluation device according to an embodiment of the present invention;

[0014] Figure 5 is a schematic diagram of an optional translation evaluation device according to an embodiment of the present application Figure 1 ;

[0015] Figure 6 is a schematic diagram of an optional translation evaluation device according to an embodiment of the present application Figure 2 ;

[0016] Figure 7 is a schematic diagram of an optional translation evaluation device according to an embodiment of the present application Figure 3 ;

[0017] Figure 8 is a schematic diagram of an optional translation evaluation device according to an embodiment of the present application Figure 4 ;

[0018] Fig. 9 is a schematic diagram of an optional translation evaluation device according to an embodiment of the present application Figure 5 ;

[0019] Fig.10 is a schematic diagram of an optional translation evaluation device according to an embodiment of the present application Figure 6 ;

[0020] Fig.11 is a schematic diagram of an optional translation evaluation device according to an embodiment of the present application Figure 7 ;as well as

[0021] Fig.12 It is a structural block diagram of a computer terminal according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0023] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0024] First, some nouns or terms that appear in the description of the embodiments of the present application are subject to the following explanations:

[0025] Cross-language search system: A search system that solves the problem that the language used by the user when searching is different from the language used by the search system.

[0026] Machine translation: also known as automatic translation, is the process of using a computer to convert one natural source language into another natural target language. It generally refers to the translation of sentences and full texts between natural languages.

[0027] Statistical machine translation: Use statistical methods to analyze a large amount of parallel corpus, build a machine translation model, and then use the model to complete the process of converting one natural language into another natural language.

[0028] Neural network translation: Use neural networks to directly model the source sentence and the target translation end-to-end, integrating all the above sub-problems into a deep neural network to solve them together.

[0029] Domain adaptation: This is a field related to machine learning and transfer learning. Learning a model from the source data distribution allows the model to have better analysis results on the target data distribution.

[0030] Example 1

[0031] According to an embodiment of the present invention, a translation optimization method embodiment is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0032] The method embodiment provided in the first embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1FIG. 1 shows a hardware structure block diagram of a computer terminal (or mobile device) for implementing a translation optimization method. Figure 1 As shown, the computer terminal 10 (or mobile device 10) may include one or more (102a, 102b, ..., 102n are used to illustrate) processors 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), and a memory 104 for storing data. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply and / or a camera. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0033] It should be noted that the one or more processors 102 and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10 (or mobile device). As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0034] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the translation optimization method in the embodiment of the present invention. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the translation optimization method of the above-mentioned application is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0035] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 (or mobile device).

[0036] Under the above operating environment, this application provides Figure 2 The translation optimization method shown. Figure 2 It is a flowchart of a translation optimization method according to Embodiment 1 of the present invention.

[0037] Step 21, obtaining a plurality of target query sentences and translation results of a plurality of target query sentences, wherein the target query sentence is a query sentence satisfying a preset condition, and the preset condition is used to characterize the translation accuracy of the query sentence in the target domain.

[0038] Specifically, when performing a query in a cross-language information query system, a large number of query statements and translation results corresponding to the query statements are generated. These online generated data can be recorded in the user behavior log, and query statements that meet preset conditions can be obtained from the user behavior log.

[0039] It should be noted that since different cross-language information query systems belong to different fields, in the process of translating query statements to query based on translation results, the translation system lacks domain data, resulting in mismatch between the translation results and the target domain, which leads to a large difference between the query results and the user's expected results. Therefore, it is necessary to obtain query statements with high translation accuracy in the target domain, and the preset condition is the translation accuracy condition of the query statement in the target domain.

[0040] For example, the target domain may be the e-commerce domain. A large number of users enter query statements on cross-language e-commerce websites every day to perform a series of actions such as searching, browsing, clicking, and purchasing. The generated behavioral data will be recorded in the log system. The domain data that is effective for translation can be mined from the large amount of user behavior data in the log. Specifically, the user's feedback data on the query results can be used to determine whether the translation results of the query statements meet the translation accuracy in the target domain, and query statements and translation results that meet the translation accuracy in the target domain can be obtained, thereby obtaining high-quality bilingual query data in the target domain. The machine translation system is optimized based on the high-quality bilingual query data to improve the accuracy of the machine translation system's translation results for query terms in the target domain.

[0041] The target query statement needs to be obtained from multiple query statements. According to the above embodiment of the present application, the above method also includes the following steps: obtaining a query statement input in a target application under a target domain, wherein the language used in the query statement is different from the language used in the target application; performing a query in the target application based on the translation result of the query statement to obtain a query result; and determining whether the query statement meets a preset condition based on the adoption rate of the query result.

[0042] Specifically, in an e-commerce system whose operating language is English, after a user enters a multilingual query statement that is different from English, a search behavior is completed through the following process: the multilingual query statement is translated by a machine translation system to obtain an English translation result. At this time, the search engine searches the English product index library based on the English translation result, obtains a series of related products, and returns the language version corresponding to the multilingual query statement to the user for the user to browse and decide whether to click on the relevant product to read detailed information, and then decide whether to buy it.

[0043] Based on the above process, the metadata is defined in the user behavior data as follows: d =<u,q,t,c> , where u represents the user, q represents the query statement, t represents the translation result of the query statement, and c represents whether the user clicks on the query result. In a cross-language search scenario, such as on AE, the online system collects a large amount of user behavior data every day. The collected user behavior data can be converted into multiple metadata. Since the metadata contains both user information and relevant information about the query statement and the translation result, and also contains information about whether the user clicks on the query result obtained based on the translation result, when judging whether a query statement meets the preset conditions, multiple metadata related to the query statement can be obtained. Based on the information of whether the user clicks on the query result in the multiple metadata, the adoption rate of the query result is determined, and then the query statement is judged to meet the preset conditions based on the adoption rate of the query result. Specifically, when the adoption rate of the query result is high, it can be determined that the query statement meets the preset conditions, and when the adoption rate of the query result is low, it can be determined that the query statement does not meet the preset conditions.

[0044] According to the above-mentioned embodiment of the present application, determining whether a query statement meets a preset condition based on the adoption rate of the query result includes: judging whether the number of times the query statement has been input is greater than a preset number of times; if the number of times the query statement has been input is greater than a preset number of times, judging whether the adoption rate of the query result is greater than a preset value; if the adoption rate of the query result is greater than a preset value, determining that the query statement meets the preset condition.

[0045] It should be noted that there are two sub-conditions for determining whether a query statement meets the preset conditions. One sub-condition is whether the adoption rate of the query result corresponding to the query statement is greater than a preset value, and the other sub-condition is whether the number of occurrences of the query statement is greater than a preset number.

[0046] Specifically, for the sub-condition of whether the number of occurrences of the query statement is greater than the preset number, in statistics, the larger the sample size, the smaller the representative error of the statistical estimate, and the smaller the sample size, the larger the estimation error. Therefore, when a query statement appears very few times, the click behavior that occurs is not credible. For example, when a query statement is only visited by one person within a period of time, that is, it is entered into the query system by one user, the behavior of this one person clicking or not clicking is very random, so the data has a lot of noise. The following formula can be used to filter out unreliable query statements and their translation results: Luv <q，t> ≥x, where Luv <q，t> represents the number of occurrences of the query statement, and χ represents the preset number of times.

[0047] For the sub-condition of whether the adoption rate of the query results corresponding to the query statement is greater than the preset value, the greater the adoption rate of the query statement and the translation result, the more reliable the translation quality of the query statement, and the smaller the adoption rate, the less reliable the translation quality of the query statement, or the search algorithm has defects, resulting in no valid search results returned in the search query system. Therefore, when the adoption rate of the query results corresponding to the query statement is greater than a certain threshold, the translation quality is relatively reliable. Through this sub-condition, unreliable query statements and their translation results can be filtered out. The specific filtering method is shown in the following formula: CTR <q，t> ≥η, where CTR <q，t> represents the adoption rate of the translation result, and η represents the preset value.

[0048] The adoption rate of the query results is determined by user behavior data. According to the above embodiment of the present application, the above method also includes the following steps: determining the adoption rate of the query results based on the number of times the user performs the target operation on the query results, wherein the target operation includes at least a click operation.

[0049] It should be noted that if after entering a multilingual query statement, the user browses the relevant query results pushed by the system, and a large number of users perform the target operation, such as clicking, it indicates that the query result is the information content expected by the user, and further indicates that the translation result truly expresses the user's meaning, and the translation quality of the multilingual query statement is relatively high. Therefore, the user adoption rate can be characterized by the conversion rate of the query statement and the translation result being clicked by the user.

[0050] Specifically, the conversion rate of query statements and translation results clicked by users is shown in the following formula: in, It indicates the total number of users who input query statement q on the search page and the total number of users who browsed the translation result t; It represents the total number of users who clicked on the query result push page. If the conversion rate of the query statement and the translation result clicked by the user satisfies the following formula, CTR <q，t> ≥η, indicating that users have a high adoption rate of the translation results of the query statement.

[0051] It should be noted that the target operation is not limited to a click operation, but can also be a browsing operation. For an e-commerce system, it can also be a purchase operation. For a document query system, it can also be a download operation. The embodiment of the present application can determine the adoption rate of the query results based on the number of executions of different target operations.

[0052] Step 22: training a target machine translation system based at least on the multiple target query sentences and the translation results of the multiple target query sentences.

[0053] Specifically, since the existing machine translation system, in the absence of domain data, obtains query results with poor adaptability to the domain after translating the query statements, in this embodiment, multiple target query statements and the translation results of the multiple target query statements constitute high-quality bilingual query data in the target domain. The high-quality bilingual query data constitutes the domain data in the target domain. The target machine translation system can be retrained based on the high-quality bilingual query data, and the existing target machine translation system can also be optimized and adjusted based on the high-quality bilingual query data, thereby improving the translation accuracy of the target machine translation system in the target domain.

[0054] According to the above-mentioned embodiment of the present application, training a target machine translation system based at least on multiple target query sentences and the translation results of the multiple target query sentences includes: determining multiple first training data based on the multiple target query sentences and the translation results of the multiple target query sentences, wherein each first training data consists of a target query sentence and the translation result corresponding to the target query sentence; training the target machine translation system based on the multiple first training data and the multiple second training data, wherein the multiple second training data are basic training data used when training machine translation systems in multiple fields.

[0055] It should be noted that the query statements and translation results that meet the two sub-conditions in the preset query condition are mined to form high-quality query statements and translated bilingual data in the target domain, that is, the first training data. When training the target machine translation system, in addition to the query statements and translated bilingual data containing domain characteristics, based on the consideration of data scale and data volume, a second training data for training the machine translation system is also required for support. Combining the first training data and the second training data, a machine translation system that is more suitable for improving the translation accuracy of query terms in the target domain can be trained, wherein the machine translation system can be a statistical machine translation system or a neural network machine translation system, and this embodiment does not limit the type of the machine translation system.

[0056] According to the above embodiment of the present application, training a target machine translation system based on multiple first training data and multiple second training data includes: mixing the multiple first training data and the multiple second training data to obtain multiple target training data; and iteratively training the target machine translation system based on the multiple target training data.

[0057] Specifically, the machine translation system can be a neural network machine translation system, which can use high-quality query sentences and translation bilingual data in the target field, that is, the first training data, to perform data enhancement on the neural network machine translation system. Specifically, the general second training data and the mined first training data are merged and randomly scattered to form a new training data. Starting from the beginning, the neural network machine translation system is iterated to obtain a neural network machine translation system with enhanced translation performance.

[0058] Through this embodiment, the mined first training data is simply and directly applied to train the machine translation system, thereby achieving the purpose of improving the machine translation performance.

[0059] According to the above embodiment of the present application, training a target machine translation system based on multiple first training data and multiple second training data includes: training a machine translation system based on multiple second training data to obtain a basic machine translation system; training a basic machine translation system based on multiple first training data to obtain a target machine translation system.

[0060] Specifically, the general second training data is first used to iterate the neural network machine translation system to obtain a basic system. Then, the first training data in the mined high-quality domain is used to continue iteratively train the basic system, and finally an optimized machine translation system is obtained.

[0061] Through this embodiment, different basic machine translation systems can be trained by mining bilingual query data in the target field, thereby simply and efficiently improving the translation accuracy of different machine translation systems for query sentences in the target field.

[0062] Step 23: Translate the query to be translated in the target domain based on the target machine translation system.

[0063] Specifically, the optimized machine translation system is used to translate the query statements to be translated, and the translation accuracy in the target domain is continuously improved, thereby improving the availability of the query results of the target query system.

[0064] It should be noted that in order to verify whether user behavior data can enhance the machine translation system, whether it can improve the translation accuracy of the query, and whether it can further improve the performance of the cross-language query system, this embodiment uses a small-scale out-of-domain data set and a large-scale in-domain data set as the data basis for verification. Among them, WMT (Workshop on Machine Translation) represents WMT2016 data, that is, out-of-domain data, DIANSHENG represents e-commerce data, that is, in-domain data, UG represents data with good user behavior, that is, data with high user positive feedback, indicating that the multilingual query sentence input by the user can effectively express the user's intention after translation, and can return relevant documents to the user, UB represents data with poor user behavior, that is, data with poor user feedback, and data with no click behavior without the user seeing the relevant documents, DA represents the data augmentation method, and FT represents the fine tuning method.

[0065] When using the above data and methods for verification, different types of data and methods can be combined. For example, the WMT+UG+DA system means that the training set uses WMT2016, the user behavior data is good, and the method uses the data enhancement method to train the system. The translation evaluation experimental results of the systems obtained by using different combinations are shown in Table 1:

[0066] Table 1

[0067]

[0068] From the coverage of the test set in the table, it can be seen that the user behavior data selected from the online system can effectively improve the coverage of the test set words. On the one hand, data with good user behavior is better than data with poor user behavior. Even if the data with good user behavior has a higher coverage in the test set than the data with poor user behavior, it can more effectively express the user's intention. On the other hand, data with poor user behavior can also improve the performance of the search query system. Through analysis, it can be seen that when problems occur in the translation or retrieval of candidate words, data with poor user behavior is generated. However, when problems occur in the retrieval algorithm, the filtered data with poor user behavior also includes translation accuracy data. In addition, it can be seen from the BLEU-4 score that the DA and FT methods perform differently on different domain data sets. On out-of-domain data, the DA method is better than the FT method, and on in-domain data, the FT method is better than the DA method.

[0069] In summary, through the data in the table, we can draw the conclusion that user behavior data can improve the translation quality of the neural network machine translation system. Specifically, whether incorporating the mined user behavior data into large-scale data within the field or small-scale data outside the field, it can obviously improve the translation quality and the search quality of the query system. Whether using the DA method or the FT method, it can also improve the translation quality and the search quality of the query system. In addition, it should be noted that when using the selected user behavior data to optimize the machine translation system, different methods can be selected according to actual conditions.

[0070] According to the above embodiment of the present application, the above method also includes the following steps: after translating the query statement to be translated in the target domain based on the target machine translation system, judging whether the query statement to be translated is a target query statement based on the adoption rate of the query result corresponding to the query statement to be translated; when the query statement to be translated is the target query statement, determining the query statement to be translated and the translation result corresponding to the query statement to be translated as the first training data.

[0071] It should be noted that in order to further optimize the translation performance of the target machine translation system, in addition to collecting more user behavior data online on the existing system and expanding the scale of high-quality bilingual query data in the target field, after the optimized target machine translation system is put into use, based on user feedback data on query results, more new first training data can be determined under the optimized target machine translation system, thereby continuously optimizing the target machine translation system in the process of updating and iterating the machine translation system, and further improving the performance of the query search system.

[0072] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0073] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0074] Example 2

[0075] According to an embodiment of the present invention, another translation optimization method is provided. Figure 3 It is a schematic diagram of a translation optimization method according to Embodiment 2 of the present invention.

[0076] First, the user enters a query statement in an online cross-language information query system. The language information retrieval can be a commodity purchasing system. In the case where the language information retrieval is a commodity purchasing system, the translation system corresponding to the information query system translates the query statement, searches for products based on the translation results, and returns them to the user. The user's browsing and clicking behavior data on the displayed products is recorded through a recording system, and a user behavior log is generated.

[0077] The data mining system analyzes the user behavior log to mine bilingual query statements that meet the requirements. Specifically, after a query statement is entered, if a large number of users click on the displayed products, it means that the translation system has a high translation quality for the query statement. On the contrary, if the number of clicks on the displayed products is small, it means that the translation quality is not high. Therefore, when the number of clicks is greater than a certain threshold, it indicates that the translation quality is reliable, and the query statement and the corresponding translation result can be used as a set of bilingual query statements. In addition, it should be noted that when a query statement is entered by a user less times, the generated click data is not reliable. Specifically, when a query statement is entered by a user only once, and the user clicks on the query result corresponding to the query statement, although the click rate is high, the query statement appears very few times. If the query statement and the corresponding translation result are used as bilingual query statements, a large error will occur. Therefore, only when the number of occurrences of the query statement and the number of clicks on the query result corresponding to the query statement meet the requirements, the query word is mined as a high-quality query statement and translation bilingual data.

[0078] After obtaining high-quality bilingual query statements, these data are then applied to the optimization and construction of the machine translation system. For example, they can be used in the data enhancement process. Specifically, the general training data and the mined data are merged and randomly scattered to obtain new training data, and the neural network machine translation system is iterated with the new training data to obtain a new machine translation system. For another example, the obtained data can also be used to fine-tune the machine translation system. Specifically, the neural network machine translation system is first iterated with general training data to obtain a basic online system, and then the mined data is used to continue training the basic online system to obtain a new machine translation system. After obtaining the new machine translation system, the query statement is translated using the system to obtain a translation result that is more suitable for the field of information query systems.

[0079] Example 3

[0080] According to an embodiment of the present invention, a device for implementing the above translation optimization method is also provided. Figure 4 As shown, the device includes: a first acquisition unit 41, a training unit 42 and a translation unit 43.

[0081] Specifically, the first acquisition unit 41 is used to acquire multiple target query sentences and translation results of multiple target query sentences, wherein the target query sentence is a query sentence that meets a preset condition, and the preset condition is used to characterize the translation accuracy of the query sentence in the target domain.

[0082] Specifically, when performing a query in a cross-language information query system, a large number of query statements and translation results corresponding to the query statements are generated. These online generated data can be recorded in the user behavior log, and query statements that meet preset conditions can be obtained from the user behavior log.

[0083] It should be noted that since different cross-language information query systems belong to different fields, in the process of translating query statements to query based on translation results, the translation system lacks domain data, resulting in mismatch between the translation results and the target domain, which leads to a large difference between the query results and the user's expected results. Therefore, it is necessary to obtain query statements with high translation accuracy in the target domain, and the preset condition is the translation accuracy condition of the query statement in the target domain.

[0084] For example, the target domain may be the e-commerce domain. A large number of users enter query statements on cross-language e-commerce websites every day to perform a series of actions such as searching, browsing, clicking, and purchasing. The generated behavioral data will be recorded in the log system. The domain data that is effective for translation can be mined from the large amount of user behavior data in the log. Specifically, the user's feedback data on the query results can be used to determine whether the translation results of the query statements meet the translation accuracy in the target domain, and query statements and translation results that meet the translation accuracy in the target domain can be obtained, thereby obtaining high-quality bilingual query data in the target domain. The machine translation system is optimized based on the high-quality bilingual query data to improve the accuracy of the machine translation system's translation results for query terms in the target domain.

[0085] The training unit 42 is configured to train a target machine translation system based at least on the multiple target query sentences and the translation results of the multiple target query sentences.

[0086] Specifically, since the existing machine translation system, in the absence of domain data, obtains query results with poor adaptability to the domain after translating the query statements, in this embodiment, multiple target query statements and the translation results of the multiple target query statements constitute high-quality bilingual query data in the target domain. The high-quality bilingual query data constitutes the domain data in the target domain. The target machine translation system can be retrained based on the high-quality bilingual query data, and the existing target machine translation system can also be optimized and adjusted based on the high-quality bilingual query data, thereby improving the translation accuracy of the target machine translation system in the target domain.

[0087] The translation unit 43 is used to translate the query sentence to be translated in the target domain based on the target machine translation system.

[0088] Specifically, the optimized machine translation system is used to translate the query statements to be translated, and the translation accuracy in the target domain is continuously improved, thereby improving the availability of the query results of the target query system.

[0089] It should be noted that in order to verify whether user behavior data can enhance the machine translation system, whether it can improve the translation accuracy of the query, and whether it can further improve the performance of the cross-language query system, this embodiment uses a small-scale out-of-domain data set and a large-scale in-domain data set as the data basis for verification. Among them, WMT (Workshop on Machine Translation) represents WMT2016 data, that is, out-of-domain data, DIANSHENG represents e-commerce data, that is, in-domain data, UG represents data with good user behavior, that is, data with high user positive feedback, indicating that the multilingual query sentence input by the user can effectively express the user's intention after translation, and can return relevant documents to the user, UB represents data with poor user behavior, that is, data with poor user feedback, and data with no click behavior without the user seeing the relevant documents, DA represents the data augmentation method, and FT represents the fine tuning method.

[0090] When using the above data and methods for verification, different types of data and methods can be combined. For example, the WMT+UG+DA system means that the training set uses WMT2016, the user behavior data is good, and the method uses the data enhancement method to train the system. The translation evaluation experimental results of the systems obtained by using different combinations are shown in Table 1:

[0091] Table 1

[0092]

[0093] From the coverage of the test set in the table, it can be seen that the user behavior data selected from the online system can effectively improve the coverage of the test set words. On the one hand, data with good user behavior is better than data with poor user behavior. Even if the data with good user behavior has a higher coverage in the test set than the data with poor user behavior, it can more effectively express the user's intention. On the other hand, data with poor user behavior can also improve the performance of the search query system. Through analysis, it can be seen that when problems occur in the translation or retrieval of candidate words, data with poor user behavior is generated. However, when problems occur in the retrieval algorithm, the filtered data with poor user behavior also includes translation accuracy data. In addition, it can be seen from the BLEU-4 score that the DA and FT methods perform differently on different domain data sets. On out-of-domain data, the DA method is better than the FT method, and on in-domain data, the FT method is better than the DA method.

[0094] In summary, through the data in the table, we can draw the conclusion that user behavior data can improve the translation quality of the neural network machine translation system. Specifically, whether incorporating the mined user behavior data into large-scale data within the field or small-scale data outside the field, it can obviously improve the translation quality and the search quality of the query system. Whether using the DA method or the FT method, it can also improve the translation quality and the search quality of the query system. In addition, it should be noted that when using the selected user behavior data to optimize the machine translation system, different methods can be selected according to actual conditions.

[0095] It should be noted that the first acquisition unit 41, the training unit 42 and the translation unit 43 correspond to steps S21 to S23 in Example 1, and the examples and application scenarios implemented by the modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the modules as part of the device can be run in the computer terminal 10 provided in Example 1.

[0096] According to the above embodiments of the present application, Figure 5 As shown, the above device further includes: a second acquiring unit 51 , a querying unit 52 and a first determining unit 53 .

[0097] Specifically, the second acquisition unit 51 is used to acquire a query statement input in a target application under a target domain, wherein the language adopted by the query statement is different from the language adopted by the target application; the query unit 52 is used to perform a query in the target application based on the translation result of the query statement to obtain a query result; the first determination unit 53 is used to determine whether the query statement meets a preset condition based on the adoption rate of the query result.

[0098] Specifically, in an e-commerce system whose operating language is English, after a user enters a multilingual query statement that is different from English, a search behavior is completed through the following process: the multilingual query statement is translated by a machine translation system to obtain an English translation result. At this time, the search engine searches the English product index library based on the English translation result, obtains a series of related products, and returns the language version corresponding to the multilingual query statement to the user for the user to browse and decide whether to click on the relevant product to read detailed information, and then decide whether to buy it.

[0099] Based on the above process, the metadata is defined in the user behavior data as follows: d =<u,q,t,c> , where u represents the user, q represents the query statement, t represents the translation result of the query statement, and c represents whether the user clicks on the query result. In a cross-language search scenario, such as on AE, the online system collects a large amount of user behavior data every day. The collected user behavior data can be converted into multiple metadata. Since the metadata contains both user information and relevant information about the query statement and the translation result, and also contains information about whether the user clicks on the query result obtained based on the translation result, when judging whether a query statement meets the preset conditions, multiple metadata related to the query statement can be obtained. Based on the information of whether the user clicks on the query result in the multiple metadata, the adoption rate of the query result is determined, and then the query statement is judged to meet the preset conditions based on the adoption rate of the query result. Specifically, when the adoption rate of the query result is high, it can be determined that the query statement meets the preset conditions, and when the adoption rate of the query result is low, it can be determined that the query statement does not meet the preset conditions.

[0100] It should be noted that the second acquisition unit 51, the query unit 52 and the first determination unit 53 correspond to the steps in Example 1, and the examples and application scenarios implemented by the modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the modules as part of the device can be run in the computer terminal 10 provided in Example 1.

[0101] According to the above embodiments of the present application, Figure 6 As shown, the query unit 52 includes: a first judgment module 521 , a second judgment module 522 and a first determination module 523 .

[0102] Specifically, the first judgment module 521 is used to judge whether the number of times the query statement is input is greater than a preset number; the second judgment module 522 is used to judge whether the adoption rate of the query result is greater than a preset value when the number of times the query statement is input is greater than a preset number; the first determination module 523 is used to determine that the query statement meets the preset conditions when the adoption rate of the query result is greater than a preset value.

[0103] It should be noted that there are two sub-conditions for determining whether a query statement meets the preset conditions. One sub-condition is whether the adoption rate of the query result corresponding to the query statement is greater than a preset value, and the other sub-condition is whether the number of occurrences of the query statement is greater than a preset number.

[0104] Specifically, for the sub-condition of whether the number of occurrences of the query statement is greater than the preset number, in statistics, the larger the sample size, the smaller the representative error of the statistical estimate, and the smaller the sample size, the larger the estimation error. Therefore, when a query statement appears very few times, the click behavior that occurs is not credible. For example, when a query statement is only visited by one person within a period of time, that is, it is entered into the query system by one user, the behavior of this one person clicking or not clicking is very random, so the data has a lot of noise. The following formula can be used to filter out unreliable query statements and their translation results: Luv <q，t> ≥χ, where Luv <q，t> represents the number of occurrences of the query statement, and χ represents the preset number of times.

[0105] For the sub-condition of whether the adoption rate of the query results corresponding to the query statement is greater than the preset value, the greater the adoption rate of the query statement and the translation result, the more reliable the translation quality of the query statement, and the smaller the adoption rate, the less reliable the translation quality of the query statement, or the search algorithm has defects, resulting in no valid search results returned in the search query system. Therefore, when the adoption rate of the query results corresponding to the query statement is greater than a certain threshold, the translation quality is relatively reliable. Through this sub-condition, unreliable query statements and their translation results can be filtered out. The specific filtering method is shown in the following formula: CTR <q，t> ≥η, where CTR <q，t> represents the adoption rate of the translation result, and η represents the preset value.

[0106] It should be noted that the first judgment module 521, the second judgment module 522 and the first determination module 523 correspond to the steps in Example 1, and the examples and application scenarios implemented by the modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the modules as part of the device can be run in the computer terminal 10 provided in Example 1.

[0107] According to the above embodiments of the present application, Figure 7 As shown, the above device also includes: a second determining unit 71.

[0108] Specifically, the second determining unit 71 is used to determine the adoption rate of the query result based on the number of times the user performs a target operation on the query result, wherein the target operation at least includes a click operation.

[0109] It should be noted that if after entering a multilingual query statement, the user browses the relevant query results pushed by the system, and a large number of users perform the target operation, such as clicking, it indicates that the query result is the information content expected by the user, and further indicates that the translation result truly expresses the user's meaning, and the translation quality of the multilingual query statement is relatively high. Therefore, the user adoption rate can be characterized by the conversion rate of the query statement and the translation result being clicked by the user.

[0110] Specifically, the conversion rate of query statements and translation results clicked by users is shown in the following formula: in, It indicates the total number of users who input query statement q on the search page and the total number of users who browsed the translation result t; It represents the total number of users who clicked on the query result push page. If the conversion rate of the query statement and the translation result clicked by the user satisfies the following formula, CTR <q，t> ≥η, indicating that users have a high adoption rate of the translation results of the query statement.

[0111] It should be noted that the target operation is not limited to a click operation, but can also be a browsing operation. For an e-commerce system, it can also be a purchase operation. For a document query system, it can also be a download operation. The embodiment of the present application can determine the adoption rate of the query results based on the number of executions of different target operations.

[0112] It should be noted that the second determining unit 71 corresponds to the steps in Embodiment 1, and the examples and application scenarios implemented by the modules are the same as those of the corresponding steps, but are not limited to the contents disclosed in Embodiment 1. It should be noted that the modules as part of the device can be run in the computer terminal 10 provided in Embodiment 1.

[0113] According to the above embodiments of the present application, Figure 8 As shown, the training unit 42 includes: a second determination module 421 and a training module 422 .

[0114] Specifically, the second determination module 421 is used to determine multiple first training data based on multiple target query sentences and the translation results of the multiple target query sentences, wherein each first training data consists of a target query sentence and the translation result corresponding to the target query sentence; the training module 422 is used to train the target machine translation system based on the multiple first training data and the multiple second training data, wherein the multiple second training data are the basic training data used when training machine translation systems in multiple fields.

[0115] It should be noted that the query statements and translation results that meet the two sub-conditions in the preset query condition are mined to form high-quality query statements and translated bilingual data in the target domain, that is, the first training data. When training the target machine translation system, in addition to the query statements and translated bilingual data containing domain characteristics, based on the consideration of data scale and data volume, a second training data for training the machine translation system is also required for support. Combining the first training data and the second training data, a machine translation system that is more suitable for improving the translation accuracy of query terms in the target domain can be trained, wherein the machine translation system can be a statistical machine translation system or a neural network machine translation system, and this embodiment does not limit the type of the machine translation system.

[0116] It should be noted that the second determination module 421 and the training module 422 correspond to the steps in Example 1, and the examples and application scenarios implemented by the modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the modules as part of the device can be run in the computer terminal 10 provided in Example 1.

[0117] According to the above embodiments of the present application, Fig. 9 As shown, the training module 422 includes: a merging submodule 4221 and a first training submodule 4222. Specifically, the merging submodule 4221 is used to mix the multiple first training data and the multiple second training data to obtain multiple target training data; the first training submodule 4222 is used to iteratively train the target machine translation system based on the multiple target training data.

[0118] Specifically, the machine translation system can be a neural network machine translation system, which can use high-quality query sentences and translation bilingual data in the target field, that is, the first training data, to perform data enhancement on the neural network machine translation system. Specifically, the general second training data and the mined first training data are merged and randomly scattered to form a new training data. Starting from the beginning, the neural network machine translation system is iterated to obtain a neural network machine translation system with enhanced translation performance.

[0119] Through this embodiment, the mined first training data is simply and directly applied to train the machine translation system, thereby achieving the purpose of improving the machine translation performance.

[0120] It should be noted that the above-mentioned merging submodule 4221 and the first training submodule 4222 correspond to the steps in Example 1, and the examples and application scenarios implemented by the above-mentioned modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the above-mentioned modules, as part of the device, can be run in the computer terminal 10 provided in Example 1.

[0121] According to the above embodiments of the present application, Fig.10 As shown, the training module 422 includes: a second training submodule 4223 and a third training submodule 4224 .

[0122] Specifically, the second training submodule 4223 is used to train the machine translation system based on multiple second training data to obtain a basic machine translation system; the third training submodule 4224 is used to train the basic machine translation system based on multiple first training data to obtain a target machine translation system.

[0123] Specifically, the general second training data is first used to iterate the neural network machine translation system to obtain a basic system. Then, the first training data in the mined high-quality domain is used to continue iteratively train the basic system, and finally an optimized machine translation system is obtained.

[0124] Through this embodiment, different basic machine translation systems can be trained by mining bilingual query data in the target field, thereby simply and efficiently improving the translation accuracy of different machine translation systems for query sentences in the target field.

[0125] It should be noted that the second training submodule 4223 and the third training submodule 4224 correspond to the steps in Example 1, and the examples and application scenarios implemented by the modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the modules as part of the device can be run in the computer terminal 10 provided in Example 1.

[0126] According to the above embodiments of the present application, Fig.11 As shown, the above device further includes: a judging unit 111 and a third determining unit 112 .

[0127] Specifically, the judging unit 111 is used to judge whether the query statement to be translated is a target query statement based on the adoption rate of the query result corresponding to the query statement to be translated after the query statement to be translated in the target domain is translated based on the target machine translation system; the third determining unit 112 is used to determine the query statement to be translated and the translation result corresponding to the query statement to be translated as the first training data when the query statement to be translated is the target query statement.

[0128] It should be noted that in order to further optimize the translation performance of the target machine translation system, in addition to collecting more user behavior data online on the existing system and expanding the scale of high-quality bilingual query data in the target field, after the optimized target machine translation system is put into use, based on user feedback data on query results, more new first training data can be determined under the optimized target machine translation system, thereby continuously optimizing the target machine translation system in the process of updating and iterating the machine translation system, and further improving the performance of the query search system.

[0129] It should be noted that the above-mentioned judgment unit 111 and the third determination unit 112 correspond to the steps in Example 1, and the examples and application scenarios implemented by the above-mentioned modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned Example 1. It should be noted that the above-mentioned modules, as part of the device, can be run in the computer terminal 10 provided in Example 1.

[0130] Example 4

[0131] The embodiment of the present invention can provide a computer terminal, which can be any computer terminal device in a computer terminal group. Optionally, in this embodiment, the computer terminal can also be replaced by a terminal device such as a mobile terminal.

[0132] Optionally, in this embodiment, the computer terminal may be located in at least one network device among a plurality of network devices of the computer network.

[0133] In this embodiment, the above-mentioned computer terminal can execute the program code of the following steps in the translation optimization method of the application: obtaining multiple target query statements and translation results of multiple target query statements, wherein the target query statement is a query statement that meets preset conditions, and the preset conditions are used to characterize the translation accuracy of the query statement in the target domain; training a target machine translation system based at least on the multiple target query statements and the translation results of the multiple target query statements; and translating the query statements to be translated in the target domain based on the target machine translation system.

[0134] Optionally, Fig.12 is a structural block diagram of a computer terminal according to an embodiment of the present invention. Fig.12 As shown, the computer terminal A may include: one or more (only one is shown in the figure) processors, a memory, and a transmission device.

[0135] Among them, the memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the translation optimization method and device in the embodiment of the present invention. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, the above-mentioned translation optimization method is realized. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories can be connected to the terminal A via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0136] The processor can call the information and application stored in the memory through the transmission device to execute the following steps: obtain multiple target query sentences and translation results of multiple target query sentences, wherein the target query sentence is a query sentence that meets a preset condition, and the preset condition is used to characterize the translation accuracy of the query sentence in the target domain; train a target machine translation system based on at least the multiple target query sentences and the translation results of the multiple target query sentences; and translate the query sentences to be translated in the target domain based on the target machine translation system.

[0137] Optionally, the processor may also execute program code for the following steps: obtaining a query statement input in a target application under a target domain, wherein the language used in the query statement is different from the language used in the target application; performing a query in the target application based on a translation result of the query statement to obtain a query result; and determining whether the query statement meets a preset condition based on an adoption rate of the query result.

[0138] Optionally, the processor may also execute the following program code steps: determining whether the query statement has been entered more than a preset number of times; if the query statement has been entered more than a preset number of times, determining whether the adoption rate of the query result is greater than a preset value; if the adoption rate of the query result is greater than a preset value, determining that the query statement meets the preset conditions.

[0139] Optionally, the processor may further execute program code of the following steps: determining the adoption rate of the query results based on the number of times the user performs a target operation on the query results, wherein the target operation includes at least a click operation.

[0140] Optionally, the processor may also execute program code of the following steps: determining multiple first training data based on multiple target query statements and translation results of the multiple target query statements, wherein each first training data consists of a target query statement and the translation result corresponding to the target query statement; training a target machine translation system based on the multiple first training data and the multiple second training data, wherein the multiple second training data are basic training data used when training machine translation systems in multiple fields.

[0141] Optionally, the processor may also execute program code of the following steps: mixing the plurality of first training data and the plurality of second training data to obtain a plurality of target training data; and iteratively training a target machine translation system based on the plurality of target training data.

[0142] Optionally, the processor may also execute program codes of the following steps: training a machine translation system based on multiple pieces of second training data to obtain a basic machine translation system; training the basic machine translation system based on multiple pieces of first training data to obtain a target machine translation system.

[0143] Optionally, the processor may also execute the program code of the following steps: after translating the query statement to be translated in the target domain based on the target machine translation system, judging whether the query statement to be translated is a target query statement based on the adoption rate of the query result corresponding to the query statement to be translated; if the query statement to be translated is the target query statement, determining the query statement to be translated and the translation result corresponding to the query statement to be translated as the first training data.

[0144] The embodiment of the present invention provides a translation optimization method. The method uses the feedback data of the query statement that meets the translation accuracy in the target field to train the machine translation system, obtains the target query statement and its translation result, and uses the target query statement and its translation result as training data to train the target translation system, and then translates the query statement through the target system, so as to achieve the purpose of improving the translation accuracy of the query statement when performing cross-language query in a specific field, thereby achieving the technical effect of improving the adoption rate of the query result, and further solves the technical problem in the related art that the low translation accuracy of the query statement leads to poor adoption of the query result when performing cross-language query in a specific field.

[0145] Those skilled in the art will appreciate that the structure shown in the figure is for illustration only, and the computer terminal may also be a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a handheld computer, a mobile Internet device (MID), a PAD, or other terminal devices. Fig.10The structure of the electronic device is not limited. For example, the computer terminal 10 may also include Fig.10 More or fewer components (such as network interfaces, display devices, etc.) shown in, or having Fig.10 Different configurations are shown.

[0146] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0147] Example 5

[0148] The embodiment of the present invention further provides a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the translation optimization method provided in the first embodiment.

[0149] Optionally, in this embodiment, the above storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0150] Optionally, in this embodiment, the storage medium is configured to store program code for executing the following steps: obtaining multiple target query statements and translation results of multiple target query statements, wherein the target query statement is a query statement that satisfies a preset condition, and the preset condition is used to characterize the translation accuracy of the query statement in the target domain; training a target machine translation system based at least on the multiple target query statements and the translation results of the multiple target query statements; and translating the query statements to be translated in the target domain based on the target machine translation system.

[0151] The translation optimization method also includes: obtaining a query statement input in a target application under a target domain, wherein the language used by the query statement is different from the language used by the target application; performing a query in the target application based on a translation result of the query statement to obtain a query result; and determining whether the query statement meets a preset condition based on an adoption rate of the query result.

[0152] Determining whether a query statement meets a preset condition based on the adoption rate of the query results includes: determining whether the query statement has been inputted more than a preset number of times; if the query statement has been inputted more than a preset number of times, determining whether the adoption rate of the query result is greater than a preset value; if the adoption rate of the query result is greater than a preset value, determining that the query statement meets the preset condition.

[0153] The translation optimization method further includes: determining the adoption rate of the query result based on the number of times the user performs a target operation on the query result, wherein the target operation at least includes a click operation.

[0154] At least based on multiple target query sentences and translation results of the multiple target query sentences, training a target machine translation system includes: determining multiple first training data based on the multiple target query sentences and the translation results of the multiple target query sentences, wherein each first training data consists of a target query sentence and the translation result corresponding to the target query sentence; training the target machine translation system based on the multiple first training data and the multiple second training data, wherein the multiple second training data are basic training data used when training machine translation systems in multiple fields.

[0155] Training a target machine translation system based on multiple first training data and multiple second training data includes: mixing the multiple first training data and the multiple second training data to obtain multiple target training data; and iteratively training the target machine translation system based on the multiple target training data.

[0156] Training a target machine translation system based on multiple first training data and multiple second training data includes: training a machine translation system based on multiple second training data to obtain a basic machine translation system; and training a basic machine translation system based on multiple first training data to obtain a target machine translation system.

[0157] The translation optimization method also includes: after translating the query statement to be translated in the target domain based on the target machine translation system, judging whether the query statement to be translated is a target query statement based on the adoption rate of the query result corresponding to the query statement to be translated; if the query statement to be translated is the target query statement, determining the query statement to be translated and the translation result corresponding to the query statement to be translated as the first training data.

[0158] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0159] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0160] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0161] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0162] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0163] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0164] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A translation optimization method, comprising: Acquire multiple target query statements and translation results of the multiple target query statements, wherein the target query statement is a query statement that meets a preset condition, and the preset condition is used to characterize the translation accuracy of the query statement in the target domain; Training a target machine translation system based at least on the multiple target query sentences and translation results of the multiple target query sentences; Translating the query to be translated in the target domain based on the target machine translation system; Wherein, the translation optimization method further comprises: Acquire a query statement input in a target application under the target domain, wherein a language adopted by the query statement is different from a language adopted by the target application; Performing a query in the target application based on the translation result of the query sentence to obtain a query result; The query sentences whose adoption rate of the query results is greater than a preset value and the translation results corresponding to the query sentences are used as training data to train the target machine translation system.

2. The translation optimization method according to claim 1, wherein: The translation optimization method further comprises: Whether the query statement meets a preset condition is determined based on the adoption rate of the query result.

3. The translation optimization method according to claim 2, wherein: Determining whether the query statement meets a preset condition based on the adoption rate of the query result includes: Determining whether the query statement has been input a number of times greater than a preset number; When the query statement has been inputted more than a preset number of times, determining whether the adoption rate of the query result is greater than a preset value; When the adoption rate of the query result is greater than a preset value, it is determined that the query statement meets the preset condition.

4. The translation optimization method according to claim 2, wherein: The translation optimization method further comprises: The adoption rate of the query result is determined based on the number of times the user performs a target operation on the query result, wherein the target operation at least includes a click operation.

5. The translation optimization method according to claim 1, wherein: Based on the multiple target query sentences and the translation results of the multiple target query sentences, training the target machine translation system includes: Determining a plurality of first training data based on the plurality of target query sentences and the translation results of the plurality of target query sentences, wherein each of the first training data consists of a target query sentence and the translation result corresponding to the target query sentence; The target machine translation system is trained based on the plurality of first training data and the plurality of second training data, wherein the plurality of second training data are basic training data used when training machine translation systems in multiple fields.

6. The translation optimization method according to claim 5, wherein: Training the target machine translation system based on the plurality of first training data and the plurality of second training data includes: Mixing the plurality of first training data and the plurality of second training data to obtain a plurality of target training data; The target machine translation system is iteratively trained based on the multiple pieces of target training data.

7. The translation optimization method according to claim 5, wherein: Training the target machine translation system based on the plurality of first training data and the plurality of second training data includes: Training a machine translation system based on the plurality of second training data to obtain a basic machine translation system; The basic machine translation system is trained based on the plurality of first training data to obtain the target machine translation system.

8. The translation optimization method according to claim 5, wherein: The translation optimization method further comprises: After translating the query to be translated in the target domain based on the target machine translation system, judging whether the query to be translated is the target query based on the adoption rate of the query result corresponding to the query to be translated; In a case where the query statement to be translated is the target query statement, the query statement to be translated and a translation result corresponding to the query statement to be translated are determined as the first training data.

9. A translation optimization device, wherein: include: A first acquisition unit is used to acquire a plurality of target query statements and translation results of the plurality of target query statements, wherein the target query statement is a query statement that meets a preset condition, and the preset condition is used to characterize the translation accuracy of the query statement in the target domain; A training unit, configured to train a target machine translation system based on the multiple target query sentences and translation results of the multiple target query sentences; A translation unit, configured to translate the query to be translated in the target domain based on the target machine translation system; The second acquisition unit is used for inputting a query statement in a target application under the target domain, wherein the language adopted by the query statement is different from the language adopted by the target application; The query unit is used to perform a query in the target application based on the translation result of the query sentence to obtain a query result; The training unit is also used to train the target machine translation system using the query sentences whose adoption rate of the query results is greater than a preset value and the translation results corresponding to the query sentences as training data.

10. A processor, characterized in that: The processor is used to run a program, wherein the program executes the translation optimization method according to any one of claims 1 to 8 when running.

Images