A search method and related apparatus

Abstract

This application discloses a search method and related apparatus. The processing device can first identify the text type corresponding to the search text, and then use different category intent recognition methods for different text types to determine the category corresponding to the search text. Finally, based on the category, it returns the target object corresponding to the search text. Therefore, on the one hand, the category intent recognition method can better match the text features, resulting in more accurate category recognition results; on the other hand, it can overcome the limitations of text relevance, obtaining more comprehensive search results, and ultimately providing users with target objects that better match their actual search intent. This is particularly beneficial in e-commerce scenarios where the text relevance between object information and search text is often poor, significantly improving the user's search experience. This application relates to artificial intelligence technology and can be applied to related scenarios of in-vehicle IoT.

Description

Technical Field

[0001] This application relates to the field of Internet technology, and in particular to a search method and related apparatus. Background Technology

[0002] With the rapid development of internet technology, various e-commerce platforms have emerged on the internet. These e-commerce platforms can establish cooperative relationships with product providers. They can obtain relevant information about the products offered by the providers and then display this information online when users search for products, allowing users to understand the product details and purchase their desired items.

[0003] Currently, product search mainly relies on text matching. By comparing the matching degree between the user's input search text and the product title, products are sorted from highest to lowest relevance and displayed to the user.

[0004] However, due to the explosive growth in the amount of information, relying on text matching can easily lead to the inclusion of a large number of unwanted search results (products) in the search results returned to users. Furthermore, for certain text searches, simple text matching may fail to understand the user's true search intent, resulting in no search results. In short, current search methods struggle to better understand users' shopping needs and display the products they truly want to buy, impacting the search experience and the economic benefits of e-commerce platforms. Summary of the Invention

[0005] To address the aforementioned technical issues, this application provides a search method that can more accurately return objects that match the user's search intent in e-commerce scenarios, thereby improving the user's search experience.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] On one hand, embodiments of this application provide a search method, the method comprising:

[0008] Get the search text;

[0009] The search text is subjected to text type identification to obtain the target text type of the search text;

[0010] Determine the category intent recognition method corresponding to the target text type. The category intent recognition method includes one of the following: category intent prediction model prediction method, click behavior log-based recognition method, and category word rewriting method.

[0011] Based on the category intent recognition method, the search text is subjected to category intent recognition to obtain category recognition results, and the category recognition results include the target category corresponding to the search text;

[0012] The query is performed using the target category, and the target object under the target category is returned.

[0013] On one hand, embodiments of this application provide a search device, which includes an acquisition unit, a first identification unit, a first determination unit, a second identification unit, and a return unit:

[0014] The acquisition unit is used to acquire the search text;

[0015] The first identification unit is used to identify the text type of the search text to obtain the target text type of the search text;

[0016] The first determining unit is used to determine the category intent recognition method corresponding to the target text type. The category intent recognition method includes one of the following: category intent prediction model prediction method, recognition method based on click behavior log, and category word rewriting method.

[0017] The second identification unit is used to perform category intent recognition on the search text based on the category intent recognition method corresponding to the target text type to obtain a category recognition result, wherein the category recognition result includes the target category corresponding to the search text;

[0018] The return unit is used to query the target category and return the target object under the target category.

[0019] In one possible implementation, if the target text type is long-tail text, and the category intent recognition method corresponding to the target text type is a category intent prediction model prediction method, the second recognition unit is specifically used for:

[0020] The search text is input into the category intent prediction model, and the category identification result is output.

[0021] In one possible implementation, the category identification result further includes the probability value corresponding to the target category, wherein the target category includes multiple categories, and the return unit is specifically used for:

[0022] The ranking weight of the corresponding category is determined based on the probability value;

[0023] Based on the ranking weights, the target objects under the multiple categories are sorted to obtain an object sequence;

[0024] Return the target object according to the object sequence.

[0025] In one possible implementation, if the target text type is short text, and the category intent recognition method corresponding to the target text type is a click behavior log-based recognition method, the second recognition unit is specifically used for:

[0026] Obtain the click behavior logs corresponding to the search text;

[0027] Based on the click behavior log, the click category distribution is determined, and the click category distribution is used to indicate the click probability of multiple categories corresponding to the search text;

[0028] The category identification result is determined from multiple categories corresponding to the search text based on the click category distribution.

[0029] In one possible implementation, the category identification result further includes a probability value of the target category, which is determined based on the click probability. The target category includes multiple categories, and the return unit is specifically used for:

[0030] The ranking weight of the corresponding category is determined based on the probability value;

[0031] Based on the ranking weights, the target objects under the multiple categories are sorted to obtain an object sequence;

[0032] Return the target object according to the object sequence.

[0033] In one possible implementation, the device further includes a second determining unit and an execution unit:

[0034] The second determining unit is used to determine the category corresponding to the search text by utilizing the textual relevance between the search text and the title information of the object;

[0035] The execution unit is configured to execute the step of obtaining the click behavior log corresponding to the search text if the number of categories corresponding to the search text is greater than a first preset threshold.

[0036] In one possible implementation, the second identification unit is specifically used for:

[0037] Obtain synonyms and / or hypernyms and hyponyms of the search text;

[0038] Expand the search text by using synonyms and / or hyponyms of the search text;

[0039] The click behavior log is obtained based on the expanded search text.

[0040] In one possible implementation, the first identification unit is specifically used for:

[0041] The search text is segmented into words to obtain segmented terms;

[0042] Based on the segmented words, the search text is identified by text type to obtain the target text type of the search text.

[0043] In one possible implementation, the first identification unit is specifically used for:

[0044] If the number of segmented words is greater than a preset threshold, the target text type is determined to be long-tail text;

[0045] If the number of segmented words is less than or equal to the preset quantity threshold, the target text type is determined to be short text.

[0046] In one possible implementation, the first identification unit is specifically used for:

[0047] If it is determined that the search text has a corresponding target category word, the target text type is determined to be category word text;

[0048] The category intent recognition method is a category word rewriting method, and the second recognition unit is specifically used for:

[0049] The category mapped to the target category term is taken as the target category.

[0050] In one possible implementation, the first identification unit is specifically used for:

[0051] The search text is segmented into words to obtain segmented terms;

[0052] If it is determined that the target word in the segmented words matches the category mapping relationship, the category word corresponding to the target word is determined as the target category word. The category mapping relationship is the mapping relationship between the target word and the category. The target word is the category word and the similar word of the category word.

[0053] or,

[0054] Calculate the similarity probability between each category term and the search text;

[0055] If the similarity probability between the target category term and the search text reaches a second preset threshold, it is determined that the search text has a corresponding target category term.

[0056] On one hand, embodiments of this application provide a computer device for searching, the computer device including a processor and a memory:

[0057] The memory is used to store program code and transmit the program code to the processor;

[0058] The processor is configured to execute the search method described above according to instructions in the program code.

[0059] In one aspect, embodiments of this application provide a computer-readable storage medium for storing program code for executing the search method described above.

[0060] On one hand, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the search method described in the foregoing aspects.

[0061] As can be seen from the above technical solution, when a user performs an object query based on search text, they can select an appropriate category intent recognition method based on different text features. This category intent recognition method can include one of the following: a category intent prediction model based on the search text itself, a category word rewriting method, or a user-dimensional recognition method based on click behavior logs. This allows for accurate category recognition even in scenarios where there is a low textual correlation between object information and search text, such as e-commerce. Thus, when determining the target text type of the search text, the category recognition result can be obtained based on the category intent recognition method corresponding to the target text type. The category recognition result includes the target category corresponding to the search text. Because a category intent recognition method that conforms to text features is used, the overall intent recognition accuracy is greatly improved, leading to a better understanding of the user's search needs. Since the target category is obtained by identifying the user's search intent, it reflects the category to which the user's desired search object belongs and reflects the user's search needs. Therefore, after identifying the target category, querying through the target category returns the target object under the target category. This achieves the return of the target object with a better understanding of the user's search needs, avoiding the lack of semantic information caused by relying on text relevance search, improving the user's online search experience, enhancing the user's shopping experience on e-commerce platforms, and also improving the economic benefits of e-commerce platforms. Attached Figure Description

[0062] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0063] Figure 1 A schematic diagram illustrating a search method in a practical application scenario provided by an embodiment of this application;

[0064] Figure 2 A flowchart illustrating a search method provided in an embodiment of this application;

[0065] Figure 3 A flowchart of model training provided for an embodiment of this application;

[0066] Figure 4 A schematic diagram illustrating a model output result provided in an embodiment of this application;

[0067] Figure 5 A flowchart for category intent recognition of short text search text provided in this application embodiment;

[0068] Figure 6 A flowchart illustrating a search method in a practical application scenario provided by an embodiment of this application;

[0069] Figure 7 A structural block diagram of a search device provided in an embodiment of this application;

[0070] Figure 8 A structural diagram of a terminal provided in an embodiment of this application;

[0071] Figure 9 This is a structural diagram of a server provided in an embodiment of this application. Detailed Implementation

[0072] The embodiments of this application will now be described with reference to the accompanying drawings.

[0073] Search is one of the most common methods users use to find items. For example, when a user wants to buy a product, they will search for it on e-commerce platforms. In related technologies, search results are mainly determined by the degree of text matching between the search text and the title information of the product. However, in e-commerce scenarios, due to the diverse and personalized nature of product information, it's easy to encounter problems with accurate matching between the search text and the user's search terms. This can lead to search results that fail to meet the user's search needs. For instance, when a user searches for "baby products," because this search text is a category term and doesn't specifically point to a particular product, some products that match the search text might not actually contain that search term in their titles, thus failing to be recommended to the user. Similarly, when a user searches for "water," they might only want to search for drinking water-related products. However, because there are numerous product titles containing this search text, such as fruits and aquatic products, the user may be presented with a large number of products that don't match their search intent, resulting in a poor search experience.

[0074] To address the aforementioned technical issues, this application provides a search method in which the processing device first identifies the text type corresponding to the search text, and then uses different category intent recognition methods to determine the category corresponding to the search text for different text types. Finally, based on the category, the target object corresponding to the search text is returned. Thus, for e-commerce scenarios with diverse object information, more comprehensive search results that better match the actual search intent can be provided to users, thereby improving the user's search experience.

[0075] It is understood that the search method provided in this application embodiment can be implemented by a computer device, which can be a terminal device or a server. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. Terminal devices include, but are not limited to, mobile phones, computers, smart voice interaction devices, smart home appliances, and in-vehicle terminals. Terminal devices and servers can be directly or indirectly connected via wired or wireless communication, which is not limited herein. Terminal devices can be user terminals, including but not limited to mobile phones, computers, smart voice interaction devices, smart home appliances, in-vehicle terminals, and aircraft. This application can be applied to in-vehicle scenarios, such as when users use in-vehicle terminal devices to search for online shopping.

[0076] This application relates to artificial intelligence (AI) technology. AI is the theory, methods, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.

[0077] Artificial intelligence (AI) is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning / deep learning. This application mainly focuses on machine learning and natural language processing technologies.

[0078] Natural Language Processing (NLP) is an important field within computer science and artificial intelligence. It studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language—the language people use in daily life—and thus it has a close relationship with linguistic research. NLP techniques typically include text processing, semantic understanding, machine translation, question answering, and knowledge graphs.

[0079] Machine Learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0080] This application can utilize machine learning technology to train a category intent prediction model, and use this model to identify the category intent of search text; using natural language processing technology, it can analyze synonyms, hyponyms, etc., to identify target category words or obtain click behavior logs.

[0081] To facilitate understanding of the technical solutions provided in this application, a search method provided in an embodiment of this application will be introduced below in conjunction with a practical application scenario.

[0082] See Figure 1 , Figure 1 This is a schematic diagram of a search method in a practical application scenario provided by an embodiment of this application. In this practical application scenario, the processing device can be a server 101 used for search processing.

[0083] Users can perform object searches on terminal devices (such as computers) like 102. When searching, users can input search text through terminal device 102 to find the objects they need. Server 101 can obtain the search text through terminal device 102, then analyze the text type to which the search text belongs, and use different category intent recognition methods for different text types to determine the category recognition results.

[0084] For example, in this practical application scenario, the user enters the search text "apple." This search text contains only one short word and lacks context, so it can be classified as short text. In this case, to return objects that match the user's search intent, server 101 can obtain the click behavior logs corresponding to the search text. Based on these logs, it determines the click category distribution. This distribution indicates the click probability of multiple categories corresponding to the search text, that is, the probability of the user clicking on objects under each category when searching for the search text "apple." Therefore, this click category distribution can, to some extent, reflect the user's search intent when searching for this text. Figure 1 As shown, among the six categories corresponding to the search text, server 101 can select the four with the highest click probability as the target categories corresponding to the search text. These target categories are those that best match the user's search intent. Based on these target categories, the category identification result corresponding to the search text can be determined. Then, based on the category identification result, the target object under the target category is returned to terminal device 102, thereby returning a more suitable search object to the user, meeting the user's search needs, and improving the user's search experience. Terminal device 101 can display the object returned by server 102 on the interface. As shown in the figure, the title information of the displayed object does not need to contain the search text, thus overcoming the limitation of text relevance.

[0085] Next, with reference to the accompanying drawings, we will introduce one of the search methods provided in this application.

[0086] See Figure 2 , Figure 2 A flowchart of a search method provided in this application embodiment, the method including:

[0087] S201: Obtain the search text.

[0088] Search text refers to the text used when searching for an object. This search text can be the text entered by the user, or the text selected by the user from multiple provided texts, etc., without limitation. The object refers to the feedback that the search function can provide to the user; that is, the object is the final search result obtained by the user when conducting the search. For example, when searching on an e-commerce platform, the object can be the products offered on the e-commerce platform.

[0089] S202: Perform text type identification on the search text to obtain the target text type of the search text.

[0090] As can be seen from the foregoing analysis of existing technologies, if search results are determined solely based on the matching degree between object title information and search text, in e-commerce scenarios, due to the diversity and personalization of object information, whether the search text is long or short, the returned search results will be difficult to meet the user's search needs. Therefore, to solve this technical problem, in this embodiment, instead of using the textual relevance between search text and object to determine search results, the search text is first subjected to category intent recognition to determine the target category corresponding to the search text. This target category is the category to which the object the user wants to search belongs. Then, objects under this target category can be returned to the user, thereby overcoming the limitation of textual relevance and returning objects that meet the user's needs as much as possible.

[0091] To achieve the above technical solution, firstly, the processing device needs to be able to accurately identify the target category corresponding to the search text. It is understandable that, due to different users' search habits, the search text entered by users when searching for a target may include various types. Different types of search text contain different amounts and types of information. Therefore, if all types of search text are identified based on the same category intent recognition method, the accuracy of the determined target category will be insufficient and will be difficult to match the user's actual search intent. For example, when determining the corresponding target category based on the textual information of the search text itself, it performs well for long-tail texts containing a large amount of information. However, for short texts containing less information, since short texts typically only include 1-2 words, it is difficult to accurately determine the corresponding target category directly based on the textual information of the short text.

[0092] Therefore, in order to accurately identify category intent for various types of search text, in this embodiment, the processing device can employ different category intent identification methods for different text types of search text. First, the processing device can perform text type identification on the search text to obtain the target text type of the search text; this target text type is the text type corresponding to the search text. In this application, text types can include long-tail text, short text, category term text, etc., which will be described in detail below based on different text types.

[0093] S203: Determine the category intent recognition method corresponding to the target text type.

[0094] After determining the target text type corresponding to the search text, the processing device can determine a suitable category intent recognition method for the search text of that target text type. In this application, the category intent recognition method may include one of the following: a category intent prediction model prediction method, a click behavior log-based recognition method, and a category word rewriting method. Among these, the category intent prediction model prediction method and the category word rewriting method identify category intent from the text content dimension of the search text.

[0095] Category intent prediction models predict category intent by analyzing the text content of search results. Category term rewriting methods determine category intent based on category terms found in the search results. Click behavior log-based identification identifies category intent from the perspective of user behavior, obtaining the category intent corresponding to the search results through analysis of user search behavior.

[0096] Therefore, in this application, the processing device can perform category intent recognition based on two dimensions: text content and object behavior. This can, to some extent, reduce the limitations of text relevance in category intent recognition and yield more accurate results. Thus, for scenarios such as e-commerce where there is often low text relevance between object information and search text, making it difficult to determine the user's desired object based solely on text relevance, this application can provide diverse and targeted category intent recognition methods to improve the user's search experience.

[0097] S204: Based on the category intent recognition method, the category intent of the search text is recognized to obtain the category recognition result.

[0098] Because this application determines category intent recognition based on text type, it can, to a certain extent, make the category intent recognition more closely match the text features of the search text, resulting in more accurate category recognition results. These results include the target category corresponding to the search text. For example, in Figure 1 In the application scenarios shown, for short texts like "apple" containing relatively little information, click behavior logs can be obtained to identify the target category corresponding to the search text from a user data perspective. This overcomes the constraint of the amount of text information and yields more accurate category identification results. For long-tail texts containing more information, the correlation between the text content and the category can be started from the text content itself to determine the corresponding target category, thus obtaining the category that is most closely related to the search text.

[0099] S205: Query by the target category and return the target object under the target category.

[0100] After identifying the target category, the processing device can determine the category to which the user's desired search object belongs. The processing device can then query multiple objects within that target category to obtain the target object, which is the object that best matches the user's search intent. The processing device can then return this target object, providing the user with more satisfactory search results and improving the user's search experience.

[0101] As can be seen from the above technical solution, when a user performs an object query based on search text, they can select an appropriate category intent recognition method based on different text features. This category intent recognition method can include one of the following: a category intent prediction model based on the search text itself, a category word rewriting method, or a user-dimensional recognition method based on click behavior logs. This allows for accurate category recognition even in scenarios where there is a low textual correlation between object information and search text, such as e-commerce. Thus, when determining the target text type of the search text, the category recognition result can be obtained based on the category intent recognition method corresponding to the target text type. The category recognition result includes the target category corresponding to the search text. Because a category intent recognition method that conforms to text features is used, the overall intent recognition accuracy is greatly improved, leading to a better understanding of the user's search needs. Since the target category is obtained by identifying the user's search intent, it reflects the category to which the user's desired search object belongs and reflects the user's search needs. Therefore, after identifying the target category, querying through the target category returns the target object under the target category. This achieves the return of the target object with a better understanding of the user's search needs, avoiding the lack of semantic information caused by relying on text relevance search, improving the user's online search experience, enhancing the user's shopping experience on e-commerce platforms, and also improving the economic benefits of e-commerce platforms.

[0102] Next, we will introduce specific methods for category intent recognition for search text of various text types.

[0103] Text Type 1: Long-tail Text

[0104] Long-tail text refers to text containing keywords and multiple related words. For example, it might contain a noun and several modifiers that modify that noun. In the long-tail text "Red Children's Small Bubble Heart Shower Gel," "shower gel" is the noun, and "red," "children," and "small bubbles" are all modifiers. Typically, long-tail text contains a large number of words related to the keywords the user wants to search for. It also includes contextual information closely related to those keywords, such as the modifiers for the noun mentioned above. Therefore, the text content of long-tail text itself contains sufficient information for category intent identification.

[0105] Based on this, for search text of the long-tail text type, the processing device can identify the target category corresponding to the search text from the perspective of its own text content. In one possible implementation, if the target text type is long-tail text, the category intent recognition method corresponding to this target text type can be a category intent prediction model. In this method, the processing device can input the search text into the category intent prediction model, and through this model, output the category recognition result. This category intent prediction model is used to determine the target category corresponding to the search text.

[0106] The intent prediction model for this category can be trained in the following way: For example... Figure 3 As shown, Figure 3 A flowchart for model training provided in this application embodiment includes:

[0107] S301: Get object information.

[0108] The object information mainly includes the object's title and category information. Understandably, object information is typically stored in relevant databases; for example, product information is stored in an e-commerce platform's database. The number of objects is substantial, making object information relatively easy to obtain, and the large data volume allows for more accurate model training. Furthermore, since both the object's title and the text content in long-tail text describe the object, they share a high degree of commonality in their text content. Based on this, the processing device can use the object's title information to simulate the long-tail text search text entered by the user when searching for an object. Compared to the user's search text, title information is easier to obtain and has a larger data volume, which facilitates more efficient model training, lowering the barrier to model training while ensuring its accuracy.

[0109] S302: Convert between uppercase and lowercase traditional Chinese characters.

[0110] To remove interference unrelated to the information expressed in the text during training, the processing device can first convert the object information into a uniform information format, such as converting all English letters to lowercase letters and converting traditional Chinese characters to simplified Chinese characters.

[0111] S303: Remove stop words.

[0112] Similarly, the processing device can also remove content that is irrelevant to the information expressed by the text content, such as stop words like "de", "le", and punctuation marks. Stop words refer to certain words or characters that are automatically filtered out before or after processing natural language data (or text) in information retrieval to save storage space and improve search efficiency. These words or characters are called stop words (Stop Words).

[0113] S304: Perform word segmentation on the title information.

[0114] To enable the model to more accurately learn the correspondence between the title information and the category of the object, the processing device can perform word segmentation on the title information, converting the title information into text information in units of words. For example, segmenting "Women's Clothing Autumn / Winter 2021 New Models" into "Women's Clothing / 2021 / Autumn / Winter / New Models". Thus, through word segmentation, the model can more meticulously learn the association between each word and the category, and thus obtain a more accurate recognition result when determining the target category corresponding to the search text.

[0115] S305: Model training.

[0116] The processing device can use the title information processed as described above as training samples, and the category information corresponding to the title information as sample labels to train the model, enabling the model to learn the association between the title information and the category it belongs to, thereby obtaining a category intention prediction model that can determine the target category based on the search text. For example, the training samples can be as shown in the following table:

[0117]

[0118] It should be noted that one category information refers to one category, and multiple category words in one category information are in a hierarchical relationship, that is, they refer to sub-categories under a large category, not multiple categories.

[0119] The processing device can use 90% of the obtained object information as the training set and 10% as the test set to train and test the model.

[0120] S306: Model persistence.

[0121] To continuously ensure the effectiveness of the model, the processing device can use continuously updated object information to perform real-time training and update on the model, making the model more suitable for category recognition of search text.

[0122] S307: Online real-time prediction.

[0123] Once the search text is obtained, the processing device can use the search text as input to the category intent prediction model, and obtain the category recognition result corresponding to the search text through the model.

[0124] To further improve the processing device's response speed to search behavior, the device can employ a text classification tool (Fasttext) based on a shallow neural network, which is fast in classification, as the category intent prediction model. Since this model only has good processing capabilities for classification, it has a simpler structure, requires less training time, and offers faster classification processing speed compared to other comprehensive models.

[0125] Text Type Two: Short Text

[0126] Short text refers to text containing a limited number of words, typically just one simple word, such as "flower," "milk," or "apple." Because short text contains less information, it often lacks contextual information relevant to the user's search query. Contextual information refers to information that reveals the meaning and context of the text. Therefore, for this type of text, relying solely on the text content may not be sufficient to accurately analyze the user's search intent, and consequently, to accurately determine the target category corresponding to the search text.

[0127] Therefore, when performing category intent recognition on short text search texts, the processing device can analyze the target category corresponding to the search text not from its own text content, but from the historical user data corresponding to the search text. It is understandable that when a user searches for an object using a certain search text, the objects clicked by the user in the search results can, to some extent, reflect the user's actual search intent; that is, the user's click behavior can reflect the object the user sought when searching for that search text. Based on this, in one possible implementation, if the target text type is short text, the category intent recognition method corresponding to this target text type can be a click behavior log-based recognition method.

[0128] In this identification method, the processing device first acquires click behavior logs corresponding to the search text. These logs identify the objects clicked by the user when searching using the search text in historical search behavior. Based on these logs, the processing device determines the click category distribution, which indicates the click probability for multiple categories corresponding to the search text. These categories represent the categories to which the clicked objects belong, and the click probability refers to the probability that the user clicks an object within each category when searching using the search text. Since user click behavior reflects their search needs, these click probabilities reflect the degree of user demand for the multiple categories corresponding to the search text. Therefore, the processing device can determine the category identification result from the multiple categories corresponding to the search text based on the click category distribution, making the category identification result more closely match the user's search intent when searching using the search text. For example, on the one hand, the processing device can directly identify all categories with a non-zero click probability as the target category corresponding to the search text, providing users with the most comprehensive search feedback; on the other hand, the processing device can also set a click probability threshold, identifying categories exceeding this threshold as target categories, thereby returning objects with a higher probability of being needed to the user, making it easier for the user to quickly find the object they need. For example, in Figure 1 In the actual application scenario shown, the processing device can set the click probability as 7% as the click probability threshold, thereby identifying the four categories with higher click probabilities among multiple categories as the target category corresponding to the search text.

[0129] Because the method for determining the target category has less dependence on the text itself, even if the search text of short text type contains less information, the processing device can obtain category recognition results that can well meet the user's search needs from the perspective of object behavior data.

[0130] Understandably, not all short texts can correspond to multiple categories; some short texts correspond to only one category. For example, when the search text is the name of a relatively obscure piece of equipment, the search text may only contain the word "equipment name," making it short text, but it only corresponds to the category described for that equipment. In this case, the processing device only needs to return the objects under that category to the user to provide a satisfactory search result.

[0131] Based on this, in one possible implementation, to further improve search efficiency and reduce the search processing pressure on the processing device, before obtaining the click behavior logs corresponding to the search text, the processing device can first use the textual relevance between the search text and the title information of the object to determine the category corresponding to the search text. Typically, when a search text is related to multiple categories, there is a high probability that objects with corresponding title information related to the search text exist in multiple categories. Therefore, the processing device can set a first preset threshold, which is used to determine whether the search text belongs to a multi-category short text, where multi-category short text refers to short text corresponding to a large number of categories. If the number of categories corresponding to the search text is greater than the first preset threshold, it indicates that there are many categories corresponding to the search text. In this case, it is necessary to analyze the user's click behavior to identify the categories that best match the user's search intent. At this time, the processing device can perform the step of obtaining the click behavior log corresponding to the search text. If the number of categories corresponding to the search text is less than the first preset threshold, it indicates that the category corresponding to the search text is relatively simple. In this case, the processing device can directly return the objects under the category corresponding to the search text to the user, which can better meet the user's search needs without further analysis through click behavior logs. This reduces the processing pressure on the processing device and improves search efficiency.

[0132] The table below shows the corresponding category information obtained from searching the short texts "apple" and "book" based on text relevance:

[0133]

[0134]

[0135] As can be seen from the table, both the short text search terms "apple" and "book" have multiple corresponding categories, so they can both be identified as multi-category short texts. The next step is to obtain the click behavior logs corresponding to the search texts.

[0136] Furthermore, in real-world search scenarios, some search terms may lack sufficient user data for accurate analysis. For example, when searching for books, most users might use "books" or "book" as search terms, while a smaller number might use terms like "small book" or "reading material." These terms are essentially synonyms, and users' search intentions are largely the same. However, due to user search habits, there may be a lack of relevant user behavior data for some of these terms. Similarly, the categories corresponding to hyponyms and supernyms are generally the same. Hyponyms are a branch of supernyms; for example, "xx brand mobile phone" is a supernym, while "xx brand mobile phone 12" and "xx brand mobile phone 13" are hyponyms. Among multiple terms that are both hyponyms and supernyms, some terms may be used less frequently in user searches. For instance, when "xx brand mobile phone 13" was first released, there might have been a lack of sufficient user behavior data for analysis.

[0137] Based on this, in one possible implementation, in order to return more comprehensive search results to users, when acquiring object behavior data, the processing device can not only acquire the data corresponding to the search text itself, but also expand the data of the search text with the object behavior data corresponding to other closely related texts based on the semantic relationship between the search text and other texts. Since the user search intent expressed by texts with strong semantic relationships is more similar, this method can significantly expand the object behavior data used for analysis while ensuring the accuracy of user search intent analysis, thereby improving the accuracy of category recognition results.

[0138] The processing device can acquire synonyms and / or hypernyms of the search text, then expand the search text using these synonyms and / or hypernyms. Based on the expanded search text, it obtains click behavior logs—that is, it acquires not only the click behavior logs of the search text itself, but also the click behavior logs of its synonyms and / or hypernyms, combining these into the overall click behavior log for the search text. Therefore, even if the amount of click behavior log data corresponding to the search text itself is small, effective and sufficient click behavior logs can be obtained for analysis based on this expansion method. For example, when "XX brand mobile phone 13" was first launched, by expanding the click behavior logs of "XX brand mobile phone" into the click behavior logs corresponding to "XX brand mobile phone 13," the problem of insufficient click behavior log data caused by the initial launch was solved.

[0139] Understandably, category information is typically complex and diverse, and objects under different categories may share certain commonalities. For example, the object "winter casual down jacket" could correspond to both the "men's clothing" and "women's clothing" categories. Therefore, in order to return more comprehensive search results to users, the processing device can identify multiple possible categories for the search text when determining the category recognition result, and return objects under multiple categories to the user.

[0140] The relevance between the same search term and multiple categories may vary. For example, for the search term "skincare products," the relevance to the category "women's personal care products" is generally greater than that to the category "men's personal care products." This relevance refers to the probability that a user will select an object from the corresponding category when searching for that search term, i.e., the user's demand for objects from different categories. Therefore, to provide users with a better search experience, one possible implementation is that the processing device can selectively return objects corresponding to the search term based on the user's demand for different categories.

[0141] The processing device may determine the degree of demand differently depending on the category intent recognition method. For category intent prediction models, since these models are essentially based on category classification of search text, the probability of the search text belonging to each category can be obtained during the prediction process. The processing device can then add the probability value corresponding to the target category to the category recognition result, where the target category includes multiple categories. For example... Figure 4 As shown, among the category recognition results corresponding to the search text "new winter casual down jacket", including the probability value corresponding to the target category, the category "clothing and underwear / women's clothing / down jacket" has the highest probability value, while the category "clothing and underwear / children's clothing / children's down jacket" has the lowest probability value.

[0142] Since this category intent prediction model determines the target category based on the correlation between search text and the category, a higher correlation indicates a higher degree of match between the category and the user's search intent. Based on this, the processing device can determine the ranking weight of the corresponding category according to the probability value, then rank the target objects under multiple categories according to this ranking weight to obtain an object sequence, and finally return the target objects according to the object sequence. The ranking weight is used to identify the importance of objects under a category in the object sequence. The higher the probability value of a category, the higher its ranking weight can be. During ranking, objects under that category can be placed in a higher position for the user's browsing and a greater number of objects can be displayed compared to other categories. Therefore, when returning target objects based on this object sequence, users can prioritize obtaining more objects that highly match their search intent, improving the user's search experience.

[0143] Regarding the click behavior log-based identification method, as mentioned above, click probabilities can reflect the user's demand for objects in different categories. Therefore, in one possible implementation, the processing device can determine the probability value of the target category based on the click probability, which can include multiple categories. Similarly, the processing device can determine the ranking weight of the corresponding category based on the probability value, and then sort the target objects under multiple categories according to the ranking weight to obtain an object sequence. Finally, the target objects are returned according to this object sequence. Thus, for categories with higher click probabilities, the processing device can prioritize object recommendations, placing objects under that category in a higher priority position, and increasing the proportion of objects under that category in the total number of returned objects, so that users can find the objects they need more efficiently.

[0144] The table below shows the click probability distribution determined for the two short text search texts "apple" and "book":

[0145]

[0146]

[0147] Comparing this table with the quantity percentages determined based on text relevance mentioned above reveals that while some categories may have a high percentage of objects within the total number of objects corresponding to a search term, they may not represent the categories users most need. For example, for the search term "apple," the category "fruit" accounts for approximately 3.7% of the objects, making it the category with the fewest objects with textual relevance to this search term. However, in actual user selection, this category has the highest click probability at 20.2%, indicating a high level of user demand for objects within this category. Therefore, determining the probability value for a category based on click probability can more accurately reflect users' actual needs and help improve the user's search experience.

[0148] like Figure 5 As shown, Figure 5 A flowchart for category intent recognition of short text search text provided in this application embodiment, the method including:

[0149] S501: Determine if the search text corresponds to multiple categories.

[0150] If yes, then the search text can be determined to be a multi-category short text; if not, then the search text is determined not to be a multi-category short text, and the processing device can directly return the object under the corresponding category of the search text.

[0151] S502: Determine short texts for multiple categories.

[0152] S503: Expand synonyms through word vectors.

[0153] The processing device can determine the synonyms corresponding to the search text by using the cosine similarity between word vectors.

[0154] S504: Expand hyponyms and hypernyms through knowledge graphs.

[0155] The processing device can acquire the knowledge graph corresponding to the search text, and determine the superordinate and subordinate terms corresponding to the search text through the hierarchical relationships identified in the knowledge graph.

[0156] S505: Get click behavior logs.

[0157] The processing device can obtain corresponding click behavior logs based on the search text expanded with synonyms and hypernyms.

[0158] S506: Determine the distribution of click categories.

[0159] S507: Determine the category identification results.

[0160] As mentioned above, the main difference between long-tail text and short text lies in the amount of information contained in the words. Therefore, in one possible implementation, the processing device can use the words contained in the search text as a way to determine the type of search text.

[0161] The processing device first segments the search text into words, which means dividing the search text into units of words. Then, the processing device can identify the text type of the search text based on the segmented words to determine the target text type of the search text.

[0162] There are various methods for text type identification based on word segmentation. For example, the processing device can determine the part of speech of the segmented words. If the segmented words contain both nouns and adjectives, it means that the search text is likely to have contextual information about the noun, and therefore contains a relatively large amount of information, which can be identified as long-tail text. If the segmented words only contain nouns, the search text is likely to be short text, lacking contextual information and containing less information.

[0163] In one possible implementation, the processing device can also determine the text type of the search text based on the number of directly effective segmented words. The processing device can set a preset threshold to determine whether the search text contains a large number of segmented words. If the number of segmented words is greater than the preset threshold, it indicates that the search text contains a large number of segmented words, corresponding to a large amount of information. Therefore, the target text type can be determined as long-tail text, and category intent identification can be performed from the perspective of the text content itself. If the number of segmented words is less than or equal to the preset threshold, it indicates that the search text contains a small number of segmented words, corresponding to a small amount of information. Therefore, the target text type can be determined as short text, and category intent identification can be performed from the perspective of object behavior data.

[0164] Text Type 3: Category Text

[0165] Unlike long-tail text and short text, category term text is not determined by the amount of information contained, but by the text content of the search text itself. Category term text refers to text used to represent a category, such as the name of a category. Because this type of text does not specifically point to a particular object, directly searching for objects based on the text content will likely result in the search terms not being included in the object's title information, leading to the inability to find the desired object. For example, if the search text is "baby products," the titles of objects under the category "baby products" are usually things like "XX brand pacifiers" or "diapers," not the text content "baby products." Furthermore, synonyms and hypernyms of category term text are often also category term text. For example, synonyms for "baby products" include "baby-related products," so searching using synonyms and hypernyms for category term text will also struggle to overcome the problems inherent in this text type.

[0166] In this embodiment, the processing device can perform object search based on the characteristics of category term text. Since category term text itself has a relatively direct correspondence with categories, the category matching the user's search intent can be directly determined through the category term text. Based on this, in one possible implementation, if it is determined that the search text has a corresponding target category term, the processing device can determine that the target text type is category term text. The category intent recognition method corresponding to this text type can be a category term rewriting method, that is, the processing device can directly use the category mapped by the target category term as the target category, so that when returning objects to the user based on the target category, the user can obtain objects that match their search intent.

[0167] To broaden the scope of category word recognition, the processing device can use various category word recognition methods to determine whether the search text contains the corresponding target category word. In one possible implementation, the processing device can first segment the search text into words, then determine whether any of the segmented words in the search text contain a target word that matches the category mapping relationship. Here, the category mapping relationship is the mapping relationship between the target word and a category; that is, if the search text contains the target word, it can be determined that the search text matches the category mapping relationship.

[0168] To better meet users' search needs and understand their search intent, the mapping relationship can be established not only based on category terms but also on words similar to category terms. For example, terms like "maternal and infant products," "pregnancy and baby supplies," "infant care," "maternal and infant related," and "baby products" can all map to the category "maternal and infant products." Therefore, in the category mapping relationship determined by the processing device, the target words can be category terms or similar words, thus enabling a more comprehensive identification of users' search intent for categories.

[0169] If it is determined that the target segmented word in the word segmentation matches the category mapping relationship, that is, the target segmented word is one or more of the target words, the processing device can determine the category word corresponding to the target segmented word as the target category word. For example, when the search text is "men's T-shirt", after word segmentation, we can get "men's clothing" and "T-shirt", where "men's clothing" can be determined as the category word corresponding to the "men's clothing" category. At this time, the processing device can directly obtain the objects under the "men's clothing" category, and then search through the "T-shirt" part of the search text to get the final target object and return it. The processing device can train a word vector model offline through a large-scale text corpus, then convert the category words into word vectors, and obtain similar words with high similarity to the category words through the cosine similarity between word vectors, and store the similarity, the mapping relationship between category words and categories.

[0170] In another possible implementation, the processing device can directly determine the target category words with a high degree of similarity to the search text based on text similarity. The processing device can preset a second threshold for determining target category words, and then calculate the similarity probability between each category word and the search text. If the similarity probability between a target category word and the search text reaches the second preset threshold, it indicates that the category corresponding to the search text is relatively close to the category corresponding to the target category word. In this case, the processing device can determine that the search text has a corresponding target category word, i.e., a target category word with a similarity probability reaching the second preset threshold, to match the user's search intent. The processing device can train a text similarity model, for example, training a network structure for a Siamese-based Sentence-Bidirectional Encoder Representation from Transformers (Sentence-BERT). After obtaining the search text, the search text and each category word can be fed into the pre-trained model to obtain the similarity probability (usually between 0 and 1).

[0171] It's important to note that long-tail text and short text are opposing text types, while category term text does not conflict with either. Search text can satisfy both criteria simultaneously. For example, the processing device can directly identify the target category term corresponding to some words in the search text, then identify the remaining text as both long-tail and short text, determine the corresponding target category, and finally combine these results to obtain the category identification result for the search text. For instance, for the search text "men's T-shirts," category term identification yields the target category term "men's clothing," and the short text "T-shirts" can be used to determine the corresponding target category.

[0172] To facilitate understanding of the technical solution provided in this application, the following section will provide an overall introduction to the search method provided in this application, using a practical application scenario as an example.

[0173] See Figure 6 , Figure 6 A flowchart illustrating a search method in a practical application scenario provided by an embodiment of this application, the method comprising:

[0174] S601: Get the search text.

[0175] S602: Perform text type recognition.

[0176] When the search text has a corresponding target category term, the processing device can determine that the search text is a category term text and perform the following steps:

[0177] S603: Use the category mapped to the target category term as the target category.

[0178] S604: Query by target category and return the target object under the target category.

[0179] When the search text has multiple segmented words after word segmentation, the processing device can determine that the search text is long-tail text and perform the following steps:

[0180] S605: Input the search text into the category intent prediction model and output the category recognition result.

[0181] S606: Determine the ranking weights based on the probability values ​​output by the model.

[0182] S607: Sort the target objects under multiple categories according to the sorting weight to obtain the object sequence.

[0183] S608: Returns the target object according to the object sequence.

[0184] When the search text has only one segmented word after word segmentation, the processing device can determine that the search text is short text and perform the following steps:

[0185] S609: Determine whether it is a short text with multiple categories.

[0186] If the number of categories corresponding to the search text is less than or equal to the first preset threshold, it can be determined that it is not a short text with multiple categories, and the following steps are performed:

[0187] S610: Determine the category corresponding to the search text as the target category and return the target object under the target category.

[0188] If the number of categories corresponding to the search text is determined to be greater than the first preset threshold, it can be identified as a short text with multiple categories, and the following steps are performed:

[0189] S611: Identify synonyms and hyponyms, and expand using synonyms and hyponyms.

[0190] S612: Get click behavior logs.

[0191] S613: Determine the distribution of click categories.

[0192] S614: Identify categories with a click probability greater than a preset probability threshold as target categories.

[0193] S615: Determine the probability value corresponding to the category based on the click probability.

[0194] S616: Determine the sorting weight based on the probability value corresponding to the category.

[0195] S617: Sort the target objects under multiple categories according to the sorting weight to obtain the object sequence.

[0196] S618: Returns the target object according to the object sequence.

[0197] Based on the search method provided in the above embodiments, this application also provides a search device, see [link to relevant documentation]. Figure 7 , Figure 7 This application provides a structural block diagram of a search device, which includes an acquisition unit 701, a first identification unit 702, a first determination unit 703, a second identification unit 704, and a return unit 705.

[0198] Acquisition unit 701 is used to acquire the search text;

[0199] The first identification unit 702 is used to identify the text type of the search text and obtain the target text type of the search text.

[0200] The first determining unit 703 is used to determine the category intent recognition method corresponding to the target text type. The category intent recognition method includes one of the following: category intent prediction model prediction method, recognition method based on click behavior log, and category word rewriting method.

[0201] The second identification unit 704 is used to perform category intent identification on the search text based on the category intent identification method corresponding to the target text type to obtain a category identification result, wherein the category identification result includes the target category corresponding to the search text;

[0202] The return unit 705 is used to query the target category and return the target object under the target category.

[0203] In one possible implementation, if the target text type is long-tail text, and the category intent recognition method corresponding to the target text type is a category intent prediction model prediction method, the second recognition unit 704 is specifically used for:

[0204] The search text is input into the category intent prediction model, and the category identification result is output.

[0205] In one possible implementation, the category identification result further includes the probability value corresponding to the target category, wherein the target category includes multiple categories, and the return unit 705 is specifically used for:

[0206] The ranking weight of the corresponding category is determined based on the probability value;

[0207] Based on the ranking weights, the target objects under the multiple categories are sorted to obtain an object sequence;

[0208] Return the target object according to the object sequence.

[0209] In one possible implementation, if the target text type is short text, and the category intent recognition method corresponding to the target text type is a click behavior log-based recognition method, the second recognition unit 704 is specifically used for:

[0210] Obtain the click behavior logs corresponding to the search text;

[0211] Based on the click behavior log, the click category distribution is determined, and the click category distribution is used to indicate the click probability of multiple categories corresponding to the search text;

[0212] The category identification result is determined from multiple categories corresponding to the search text based on the click category distribution.

[0213] In one possible implementation, the category identification result further includes a probability value of the target category, which is determined based on the click probability. The target category includes multiple categories, and the return unit 705 is specifically used for:

[0214] The ranking weight of the corresponding category is determined based on the probability value;

[0215] Based on the ranking weights, the target objects under the multiple categories are sorted to obtain an object sequence;

[0216] Return the target object according to the object sequence.

[0217] In one possible implementation, the device 700 further includes a second determining unit and an execution unit:

[0218] The second determining unit is used to determine the category corresponding to the search text by utilizing the textual relevance between the search text and the title information of the object;

[0219] The execution unit is configured to execute the step of obtaining the click behavior log corresponding to the search text if the number of categories corresponding to the search text is greater than a first preset threshold.

[0220] In one possible implementation, the second identification unit 704 is specifically used for:

[0221] Obtain synonyms and / or hypernyms and hyponyms of the search text;

[0222] Expand the search text by using synonyms and / or hyponyms of the search text;

[0223] The click behavior log is obtained based on the expanded search text.

[0224] In one possible implementation, the first identification unit 702 is specifically used for:

[0225] The search text is segmented into words to obtain segmented terms;

[0226] Based on the segmented words, the search text is identified by text type to obtain the target text type of the search text.

[0227] In one possible implementation, the first identification unit 702 is specifically used for:

[0228] If the number of segmented words is greater than a preset threshold, the target text type is determined to be long-tail text;

[0229] If the number of segmented words is less than or equal to the preset quantity threshold, the target text type is determined to be short text.

[0230] In one possible implementation, the first identification unit 702 is specifically used for:

[0231] If it is determined that the search text has a corresponding target category word, the target text type is determined to be category word text;

[0232] The category intent recognition method is a category word rewriting method, and the second recognition unit 704 is specifically used for:

[0233] The category mapped to the target category term is taken as the target category.

[0234] In one possible implementation, the first identification unit 702 is specifically used for:

[0235] The search text is segmented into words to obtain segmented terms;

[0236] If it is determined that the target word in the segmented words matches the category mapping relationship, the category word corresponding to the target word is determined as the target category word. The category mapping relationship is the mapping relationship between the target word and the category. The target word is the category word and the similar word of the category word.

[0237] or,

[0238] Calculate the similarity probability between each category term and the search text;

[0239] If the similarity probability between the target category term and the search text reaches a second preset threshold, it is determined that the search text has a corresponding target category term.

[0240] It is understood that in the specific implementation of this application, user information (such as user click behavior logs) and other related data are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0241] This application also provides an electronic device for searching, which can be a terminal, for example, a smartphone:

[0242] Figure 8 The diagram shown is a block diagram of a portion of the structure of a smartphone provided in an embodiment of this application. (Reference) Figure 8A smartphone includes components such as a radio frequency (RF) circuit 1110, a memory 1120, an input unit 1130, a display unit 1140, a sensor 1150, an audio circuit 1160, a Wi-Fi module 1170, a processor 1180, and a power supply 1190. The input unit 1130 may include a touch panel 1131 and other input devices 1132, the display unit 1140 may include a display panel 1141, and the audio circuit 1160 may include a speaker 1161 and a microphone 1162. It is understood that... Figure 8 The smartphone structure shown does not constitute a limitation on smartphones and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0243] The memory 1120 can be used to store software programs and modules. The processor 1180 executes various functions and data processing of the smartphone by running the software programs and modules stored in the memory 1120. The memory 1120 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, applications required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the smartphone (such as audio data, phonebook, etc.). In addition, the memory 1120 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0244] The processor 1180 is the control center of the smartphone, connecting various parts of the smartphone via various interfaces and lines. It performs various functions and processes data by running or executing software programs and / or modules stored in the memory 1120 and calling data stored in the memory 1120. Optionally, the processor 1180 may include one or more processing units; preferably, the processor 1180 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 1180.

[0245] In this embodiment, the processor 1180 in the smartphone can perform the following steps:

[0246] Get the search text;

[0247] The search text is subjected to text type identification to obtain the target text type of the search text;

[0248] Determine the category intent recognition method corresponding to the target text type. The category intent recognition method includes one of the following: category intent prediction model prediction method, click behavior log-based recognition method, and category word rewriting method.

[0249] Based on the category intent recognition method, the search text is subjected to category intent recognition to obtain category recognition results, and the category recognition results include the target category corresponding to the search text;

[0250] The query is performed using the target category, and the target object under the target category is returned.

[0251] This application also provides a server; please refer to [link / reference]. Figure 9 As shown, Figure 9 This is a structural diagram of a server 1200 provided in an embodiment of this application. The server 1200 can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 1222 (e.g., one or more processors) and a memory 1232, and one or more storage media 1230 (e.g., one or more mass storage devices) for storing application programs 1242 or data 1244. The memory 1232 and storage media 1230 can be temporary or persistent storage. The program stored in the storage media 1230 may include one or more modules (not shown in the diagram), each module including a series of instruction operations on the server. Furthermore, the CPU 1222 may be configured to communicate with the storage media 1230 and execute the series of instruction operations in the storage media 1230 on the server 1200.

[0252] Server 1200 may also include one or more power supplies 1226, one or more wired or wireless network interfaces 1250, one or more input / output interfaces 1258, and / or one or more operating systems 1241, such as Windows Server. TM Mac OS X TM Unix TM Linux TM FreeBSD TM etc.

[0253] In this embodiment, the aforementioned steps performed by the server can be based on Figure 9 The server architecture shown is implemented.

[0254] According to one aspect of this application, a computer-readable storage medium is provided for storing program code for performing the search methods described in the foregoing embodiments.

[0255] According to one aspect of this application, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the search method provided in various optional implementations of the above embodiments.

[0256] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0257] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0258] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between apparatuses or units through some interfaces, and may be electrical, mechanical, or other forms.

[0259] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0260] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0261] If the integrated unit is implemented as 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 this application, 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. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0262] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A search method, characterized in that, The method includes: Get the search text; The search text is subjected to text type identification to obtain the target text type of the search text. The text type includes long-tail text, short text and category word text. Long-tail text refers to text containing keywords and multiple words related to the keywords. Short text refers to text containing one simple word. Category word text refers to text used to represent categories. The category intent recognition method corresponding to the target text type is determined. This method includes one of the following: a category intent prediction model prediction method, a click behavior log-based recognition method, and a category word rewriting method. The category intent prediction model prediction method analyzes the text content of the search text using a model to determine the matching category intent. The category word rewriting method determines the corresponding category intent based on the category words contained in the search text. The click behavior log-based recognition method identifies category intent from the perspective of object behavior, obtaining the category intent corresponding to the search text through analysis of user search behavior. If the target text type is long-tail text, the category intent recognition method corresponding to the target text type is the category intent prediction model prediction method. If the target text type is short text, the category intent recognition method corresponding to the target text type is the click behavior log-based recognition method. If the target text type is category word text, the category intent recognition method corresponding to the target text type is the category word rewriting method. Based on the category intent recognition method, the search text is subjected to category intent recognition to obtain category recognition results, and the category recognition results include the target category corresponding to the search text; The query is performed using the target category, and the target object under the target category is returned.

2. The method according to claim 1, characterized in that, If the target text type is long-tail text, and the category intent recognition method corresponding to the target text type is the category intent prediction model prediction method, then the step of performing category intent recognition on the search text based on the category intent recognition method to obtain the category recognition result includes: The search text is input into the category intent prediction model, and the category identification result is output.

3. The method according to claim 2, characterized in that, The category identification result also includes the probability value corresponding to the target category, wherein the target category includes multiple categories, and the step of returning the target object under the target category includes: The ranking weight of the corresponding category is determined based on the probability value; Based on the ranking weights, the target objects under the multiple categories are sorted to obtain an object sequence; Return the target object according to the object sequence.

4. The method according to claim 1, characterized in that, If the target text type is short text, and the category intent recognition method corresponding to the target text type is the recognition method based on click behavior logs, then the category intent recognition of the search text based on the category intent recognition method to obtain the category recognition result includes: Obtain the click behavior logs corresponding to the search text; Based on the click behavior log, the click category distribution is determined, and the click category distribution is used to indicate the click probability of multiple categories corresponding to the search text; The category identification result is determined from multiple categories corresponding to the search text based on the click category distribution.

5. The method according to claim 4, characterized in that, The category identification result also includes the probability value of the target category, which is determined based on the click probability. The target category includes multiple categories, and returning the target object under the target category includes: The ranking weight of the corresponding category is determined based on the probability value; Based on the ranking weights, the target objects under the multiple categories are sorted to obtain an object sequence; Return the target object according to the object sequence.

6. The method according to claim 4, characterized in that, Before obtaining the click behavior logs corresponding to the search text, the method further includes: By utilizing the textual relevance between the search text and the title information of the object, the category corresponding to the search text is determined; If the number of categories corresponding to the search text is greater than a first preset threshold, the step of obtaining the click behavior log corresponding to the search text is executed.

7. The method according to claim 4, characterized in that, The step of obtaining the click behavior log corresponding to the search text includes: Obtain synonyms and / or hypernyms and hyponyms of the search text; Expand the search text by using synonyms and / or hyponyms of the search text; The click behavior log is obtained based on the expanded search text.

8. The method according to any one of claims 1-7, characterized in that, The step of performing text type identification on the search text to obtain the target text type of the search text includes: The search text is segmented into words to obtain segmented terms; Based on the segmented words, the search text is identified by text type to obtain the target text type of the search text.

9. The method according to claim 8, characterized in that, The step of identifying the target text type of the search text based on the segmented words to obtain the target text type of the search text includes: If the number of segmented words is greater than a preset threshold, the target text type is determined to be long-tail text; If the number of segmented words is less than or equal to the preset quantity threshold, the target text type is determined to be short text.

10. The method according to claim 1, characterized in that, The step of performing text type identification on the search text to obtain the target text type of the search text includes: If it is determined that the search text has a corresponding target category word, the target text type is determined to be category word text; The category intent recognition method is the category word rewriting method. The step of performing category intent recognition on the search text based on the category intent recognition method to obtain category recognition results includes: The category mapped to the target category term is taken as the target category.

11. The method according to claim 10, characterized in that, The step of determining that the search text has a corresponding target category term includes: The search text is segmented into words to obtain segmented terms; If it is determined that the target word in the segmented words matches the category mapping relationship, the category word corresponding to the target word is determined as the target category word. The category mapping relationship is the mapping relationship between the target word and the category. The target word is the category word and the similar word of the category word. or, Calculate the similarity probability between each category term and the search text; If the similarity probability between the target category term and the search text reaches a second preset threshold, it is determined that the search text has a corresponding target category term.

12. A search device, characterized in that, The device includes an acquisition unit, a first identification unit, a first determination unit, a second identification unit, and a return unit. The acquisition unit is used to acquire the search text; The first identification unit is used to identify the text type of the search text to obtain the target text type of the search text. The text type includes long-tail text, short text and category word text. The long-tail text refers to text containing keywords and multiple words related to the keywords. The short text refers to text containing one simple word. The category word text refers to text used to represent a category. The first determining unit is used to determine the category intent recognition method corresponding to the target text type. The category intent recognition method includes one of the following: a category intent prediction model prediction method, a click behavior log-based recognition method, and a category word rewriting method. The category intent prediction model prediction method is based on the analysis of the text content of the search text by a model to determine the matching category intent. The category word rewriting method is based on the category words contained in the search text to determine the corresponding category intent. The click behavior log-based recognition method is to identify the category intent from the object behavior dimension and obtain the category intent corresponding to the search text by analyzing the user's search behavior. If the target text type is long-tail text, the category intent recognition method corresponding to the target text type is the category intent prediction model prediction method. If the target text type is short text, the category intent recognition method corresponding to the target text type is the click behavior log-based recognition method. If the target text type is category word text, the category intent recognition method corresponding to the target text type is the category word rewriting method. The second identification unit is used to perform category intent identification on the search text based on the category intent identification method to obtain a category identification result, wherein the category identification result includes the target category corresponding to the search text; The return unit is used to query the target category and return the target object under the target category.

13. The apparatus according to claim 12, characterized in that, If the target text type is long-tail text, and the category intent recognition method corresponding to the target text type is the category intent prediction model prediction method, the second recognition unit is specifically used for: The search text is input into the category intent prediction model, and the category identification result is output.

14. The apparatus according to claim 13, characterized in that, The category identification result also includes the probability value corresponding to the target category. The target category includes multiple categories. The return unit is specifically used for: The ranking weight of the corresponding category is determined based on the probability value; Based on the ranking weights, the target objects under the multiple categories are sorted to obtain an object sequence; Return the target object according to the object sequence.

15. A computer device, characterized in that, The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is configured to execute the search method according to any one of claims 1-11 according to the instructions in the program code.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program for performing the search method according to any one of claims 1-11.

17. A computer program product comprising instructions that, when run on a computer, cause the computer to perform the search method according to any one of claims 1-11.

Images