Method and system for optimizing commodity titles and computing equipment

Abstract

The invention provides a method and system for optimizing a commodity title and computing equipment, and the method comprises the steps: obtaining commodity information which comprises an original title of a commodity and a commodity attribute; obtaining related flow words associated with the commodity by using an algorithm recommendation system; based on the commodity information, first candidate traffic words used for being added into the original title are selected from the related traffic words through a large language model; based on the commodity information and the first recommended traffic word, utilizing a large language model to generate a second candidate traffic word used for being added into the original title; based on the commodity information, the first candidate traffic word and the second candidate traffic word, determining excluded risk traffic words by using a large language model to obtain a third candidate traffic word; and selecting an optimized traffic word from the third candidate traffic words by utilizing a large language model based on the commodity information, and adding the optimized traffic word into the original title, thereby generating a new title. According to the technical scheme, merchants can be better helped to screen drainage keywords, commodity titles can be more conveniently optimized, and commodity competitiveness is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce, and in particular to a method and system for optimizing commodity titles and a computing device. Background Art

[0002] The quality of keywords in product titles is a key factor in whether products can be retrieved by the main search. If they are well edited, they can effectively increase product exposure without paying extra fees. Merchants who are sensitive to product materials will spend a lot of time optimizing keyword traffic. For example, they check the merchant's unified data platform every day to see which words buyers have searched for recently and which words are popular in the industry, and add them to their related products. Doing a good job of keyword traffic requires a professional operator and a lot of effort. After having a large model, automatic title optimization becomes one of the core capabilities of intelligent operations, helping merchants optimize product materials and improve product competitiveness.

[0003] To this end, a technical solution is needed that can better help merchants screen traffic-generating keywords, optimize product titles more conveniently, and improve product competitiveness. Summary of the invention

[0004] The present invention aims to provide a method and system and computing device for optimizing product titles, which can better help merchants screen traffic-generating keywords, optimize product titles more conveniently, and enhance product competitiveness.

[0005] According to one aspect of the present invention, a method for optimizing a product title is provided, the method comprising:

[0006] Obtaining product information, including the original title and product attributes of the product;

[0007] Using an algorithm recommendation system to obtain relevant traffic words associated with the product;

[0008] Selecting a first candidate traffic word for adding to the original title from the relevant traffic words using a large language model based on the product information;

[0009] Generate a second candidate traffic word for adding to the original title using a large language model based on the product information and the first candidate traffic word;

[0010] Based on the product information and the first candidate traffic word and the second candidate traffic word, a large language model is used to determine and exclude risky traffic words to obtain a third candidate traffic word;

[0011] Based on the product information, a large language model is used to select an optimized traffic word from the third candidate traffic word and the optimized traffic word is added to the original title, thereby generating a new title.

[0012] According to some embodiments, the relevant traffic words include:

[0013] The query keywords with high exposure for the products in the corresponding leaf categories within the predetermined time period;

[0014] Selling point keywords extracted from the titles of competing products;

[0015] The product’s own traffic-generating keywords.

[0016] According to some embodiments, obtaining relevant traffic words associated with the product based on a recommendation algorithm includes:

[0017] Performing a weighted semantic similarity calculation on the commodity traffic word, the category name of the commodity and the original title of the commodity;

[0018] According to the weighted calculation result, the relevant traffic words are obtained using a pre-trained small model.

[0019] According to some embodiments, a natural language processing tool is used to filter out Chinese words with a predetermined frequency from the word segmentation of the full amount of titles;

[0020] Build a maintainable custom word segmentation dictionary using the Chinese vocabulary;

[0021] The original title is split using the custom word segmentation dictionary.

[0022] According to some embodiments, the method is implemented by a plurality of agents associated with the algorithmic recommendation system and the large language model.

[0023] According to another aspect of the present invention, a system for optimizing product titles is provided, the system comprising:

[0024] A plurality of intelligent agent modules, wherein the plurality of intelligent agent modules are used for a service platform to send requests and return results, wherein the service platform includes a large language model and an algorithm recommendation system;

[0025] a partitioning module, the partitioning module being used to combine at least a portion of the plurality of agent modules;

[0026] A pipeline process processing module, the pipeline process processing module is used to set the execution order of the partition modules and set result checkpoints,

[0027] Wherein, the multiple agent modules include:

[0028] The first agent obtains product information, where the product information includes the original title and product attributes of the product;

[0029] The second agent uses an algorithm recommendation system to obtain relevant traffic words associated with the product, and selects a first candidate traffic word for adding to the original title from the relevant traffic words using a large language model based on the product information;

[0030] A third intelligent agent generates a second candidate traffic word for adding to the original title based on the product information and the first recommended traffic word using a large language model;

[0031] A fourth intelligent agent determines and excludes risky traffic words based on the commodity information, the first candidate traffic word, and the second candidate traffic word using a large language model to obtain a third candidate traffic word;

[0032] The fifth intelligent agent uses a large language model to select optimized traffic words from the third candidate traffic words based on the product information and adds the optimized traffic words to the original title, thereby generating a new title.

[0033] According to some embodiments, the plurality of agent modules are implemented in a class manner;

[0034] The partition module realizes combining at least a part of the multiple agent modules by means of dynamic instantiation.

[0035] According to some embodiments, the pipeline processing flow module is implemented by a pipeline processing flow class and stores the context of the execution process of the multiple intelligent agent modules.

[0036] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the computer program executes any of the above methods.

[0037] According to another aspect of the present invention, there is provided a computing device, comprising:

[0038] Processor; and

[0039] A memory storing a computer program, which, when executed by the processor, enables the processor to execute any of the methods described above.

[0040] According to an embodiment of the present invention, after obtaining product information, an algorithm recommendation system is used to obtain relevant traffic words associated with the product, a first candidate traffic word for adding to the original title is selected from the relevant traffic words based on the product information using a large language model, a second candidate traffic word for adding to the original title is generated based on the product information and the first candidate traffic word using a large language model, and a third candidate traffic word is obtained by determining the excluded risk traffic words based on the product information, the first candidate traffic word, and the second candidate traffic word using a large language model, and an optimized traffic word is selected from the third candidate traffic word based on the product information using a large language model, and the optimized traffic word is added to the original title, thereby generating a new title. The method of the present invention can better help merchants screen traffic-driving keywords, more conveniently optimize product titles, and enhance product competitiveness.

[0041] According to some embodiments, the method of the present invention has high accuracy and relevance, and can automatically generate highly relevant keywords or phrases based on the input product description, characteristics and other information, which helps to improve the relevance of search results and makes it easier for potential customers to find products of interest to them.

[0042] According to some embodiments, the method of the present invention has a wide vocabulary library and language understanding capabilities, and can create a diverse and comprehensive combination of traffic words for various types of goods, not only limited to common terms but also including industry-specific professional vocabulary.

[0043] According to some embodiments, completely manually creating effective key traffic words may be very time-consuming and require professional knowledge. Using the method of the present invention can complete this task faster, significantly reduce the time required, and reduce the need for professionals, thereby saving human resource costs.

[0044] According to some embodiments, market trends, consumer preferences, and search engine algorithms are constantly changing. The method of the present invention continuously updates training to adapt to the latest data patterns, ensuring that the generated content remains up to date and in line with current preferred practices.

[0045] It is to be understood that the foregoing general description and the following detailed description are exemplary only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for describing the embodiments are briefly introduced below.

[0047] Figure 1 A flow chart of a method for optimizing a product title according to an example embodiment is shown.

[0048] Figure 2 A schematic diagram illustrating a detailed optimization process according to an example embodiment.

[0049] Figure 3 A schematic diagram showing a system for optimizing product titles according to an example embodiment.

[0050] Figure 4 A schematic diagram showing a multi-agent implementation according to an example embodiment.

[0051] Figure 5 A schematic diagram showing a multi-agent design according to an example embodiment.

[0052] Figure 6 A block diagram of a computing device is shown according to an example embodiment. DETAILED DESCRIPTION

[0053] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that the present invention will be comprehensive and complete and fully convey the concepts of the example embodiments to those skilled in the art. The same reference numerals in the figures represent the same or similar parts, and thus their repeated description will be omitted.

[0054] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present invention. However, those skilled in the art will appreciate that the technical solution of the present invention can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring various aspects of the present invention.

[0055] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0056] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0057] It should be understood that although the terms first, second, third, etc. may be used herein to describe various components, these components should not be limited by these terms. These terms are used to distinguish one component from another component. Therefore, the first component discussed below can be referred to as the second component without departing from the teachings of the present inventive concept. As used herein, the term "and / or" includes any one of the associated listed items and all combinations of one or more.

[0058] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0059] Those skilled in the art will appreciate that the drawings are merely schematic diagrams of example embodiments, and the modules or processes in the drawings are not necessarily necessary for implementing the present invention, and therefore cannot be used to limit the protection scope of the present invention.

[0060] In order to quickly respond to market changes, changes in user needs, and capture business opportunities, keywords in product titles, as an important part of traffic generation, face some difficulties in screening. Choosing which keywords can best describe product characteristics and attract buyers' attention at the same time requires a deep understanding of the target market. Merchants may find it difficult to determine which are high-search volume, low-competition keyword combinations. Too much keyword stacking may make the title appear unnatural, affect the user's reading experience, and even be regarded as spam by search engines and downgraded. Therefore, it is necessary to keep the title concise and attractive while optimizing search engine rankings.

[0061] Different e-commerce platforms have different restrictions on the number of characters in product titles. How to include the most important keywords in a limited space is a challenge. Short titles need to be refined, and long titles should avoid redundancy. Whether the brand name should be emphasized in the title or a more general product description should be used also depends on the brand's popularity and the unique selling points of the product itself. For niche or new brands, the brand name alone may not be enough to attract traffic. If selling to the international market, factors such as language habits and cultural backgrounds in different regions also need to be considered to ensure that the selected keywords are effective in the local market.

[0062] Some keywords in product titles may be seasonal or change with popular trends, which requires merchants to regularly update the keywords in the title to maintain their relevance. Analyzing competitor product titles can help discover effective keyword strategies, but this also requires a lot of time and effort to research. The lack of data feedback on keyword performance makes it difficult to evaluate the effectiveness of currently used keywords and how to adjust and optimize them.

[0063] To this end, the present invention proposes a method for optimizing product titles. Through a large language model, product traffic words are screened, and traffic words in the title are added, modified or deleted to optimize the product title. The method of the present invention can better help merchants screen traffic keywords, optimize product titles more conveniently, and improve product competitiveness.

[0064] Before describing the embodiments of the present invention, some terms or concepts involved in the embodiments of the present invention are explained.

[0065] LLM: LLMs / Large Language Model / Large Language Models / Large Language Model. A large language model is a computer model that can understand and generate natural language. It uses deep learning technology and a large amount of text data for training to obtain the grammatical, semantic, and contextual information of the language. Large language models can be used for tasks such as machine translation, question-answering systems, and automatic summarization, and can generate coherent and accurate text. In recent years, with the development of deep learning technology, the performance and application areas of large language models have continued to expand.

[0066] Agent: Generally refers to an entity with a certain degree of "intelligence" or decision-making ability, which can perform corresponding operations based on external input, that is, an intelligent agent. In the field of big models and artificial intelligence, "Agent" refers to an entity that can perceive its environment and make decisions or take actions based on the environment. This term can be applied to a variety of scenarios, including but not limited to: Software agent: In computer science, a software agent is an automated program that can perform specific tasks, such as data processing, information retrieval, etc. In artificial intelligence, especially reinforcement learning and multi-agent systems, an agent is a unit that performs actions. It can learn and optimize its behavior by interacting with the environment. Conversational agent: In the field of natural language processing, intelligent conversational agents such as chatbots can interact with users, understand user intent, and generate responses.

[0067] Multi-agent systems: In some complex systems, multiple agents can collaborate or compete with each other to achieve specific goals. Such systems involve more interactions and dynamic adjustments.

[0068] COT: Chain-of-Thought "Chain-of-Thought" technology is a technology based on artificial intelligence and natural language processing, which aims to simulate the human thinking process to achieve understanding and reasoning of natural language text. It involves using complex algorithms and models to analyze the logical relationships and semantic information in the text to build a series of related thoughts or opinions. This technology can be applied to natural language understanding, dialogue systems, recommendation systems and other fields, helping computers to better understand and process human language and thinking patterns. The development of Chain-of-Thought technology can improve the accuracy and intelligence level of artificial intelligence systems when processing natural language tasks.

[0069] Merchant unified data platform: provides a one-stop, personalized, customizable business decision-making experience for all merchants. It integrates massive data and store management ideas, and can not only better provide merchants with data disclosure, analysis, interpretation, and prediction functions for the entire chain of store operations such as traffic, goods, and transactions, but also better guide merchants' data-based operations.

[0070] Exemplary embodiments of the present invention are described below with reference to the accompanying drawings.

[0071] Figure 1 A flow chart of a method for optimizing a product title according to an example embodiment is shown.

[0072] See also Figure 1 In S101, product information is obtained, where the product information includes the original title and product attributes of the product.

[0073] According to some embodiments, product attributes include product name, brand, price, category / classification, material, size, origin, etc.

[0074] An example will be described below.

[0075] Input the following information to the large language model:

[0076] Now there is an e-commerce product with the following product information: the current title of the product is [Dyed Quilted Embroidery Sequins Three-dimensional Leaf Bead Tube Embroidery Fabric Cross-border Children's Clothing Wedding Dress Mesh Fabric Mesh]; the attributes of the product include the ingredients and content [Polyester], yarn count [as per sample], gram weight [0.92 (g / ㎡)], width [125-135], embroidery craft [sequin embroidery], origin [Guangdong], specific use [stage costume], applicable season [spring and summer], pattern style [fine pattern], brand [HLE-DY], popular elements [Europe and the United States], major downstream platforms [Amazon], and major sales areas [Africa].

[0077] In S103, the algorithm recommendation system is used to obtain relevant traffic words associated with the product.

[0078] The relevant traffic words include: the commodity traffic words include query keywords with high exposure in the corresponding leaf category of the commodity within the predetermined time period; selling point keywords extracted from the titles of competing commodities; and the traffic keywords of the commodity itself. The predetermined time period is usually the past seven days.

[0079] Based on the recommendation algorithm, relevant traffic words associated with the product are obtained, and the semantic similarity of the product traffic words, the category name of the product and the original title of the product is weighted and calculated; according to the weighted calculation result, the relevant traffic words are obtained using the pre-trained small model. For example, "round neck" and "v neck" have high semantic similarity.

[0080] The open source word segmentation tool (NPL) is prone to misclassification of long words. NPL cannot correctly segment some long words. For example, "fast charging charger" will be divided into "fast", "charge", and "charger", but it should actually be "fast charging" and "charger"; "European and American style", "ready-to-eat", "no cooking" and other combination words and abbreviations are also easily misclassified during word segmentation. Through the natural language processing tool, Chinese words with a predetermined number of word frequencies are screened out from the full title segmentation; the Chinese words are used to build a maintainable custom word segmentation dictionary; the original title is split using the custom word segmentation dictionary; the custom word segmentation dictionary is used to eliminate the misclassification of long words by the open source word segmentation tool. Chinese words with a word frequency greater than 10 are screened out from the platform's natural language processing tool (alinlp) for the full title segmentation. Each appearance in a product title after segmentation is counted as one word frequency increase by 1, covering all products on the platform.

[0081] In S105, a first candidate traffic word for adding to the original title is selected from the related traffic words based on the product information using a large language model.

[0082] According to some embodiments, based on the product information and algorithm recommendation, a large language model is used to select the first candidate traffic word for adding to the original title from the relevant traffic words, and an API interface and a large language model LLM are used. When filtering traffic words, the large language model filters the product traffic words and gives reasons.

[0083] An example will be described below.

[0084] Input the following information to the large language model:

[0085] Now there is an e-commerce product with the following product information: the current title of the product is [Dyed Quilted Embroidery Sequins Three-dimensional Leaf Bead Tube Embroidery Fabric Cross-border Children's Clothing Wedding Dress Mesh Fabric Mesh]; the attributes of the product include the ingredients and content [Polyester], yarn count [as per sample], gram weight [0.92 (g / ㎡)], width [125-135], embroidery craft [sequin embroidery], origin [Guangdong], specific use [stage costume], applicable season [spring and summer], pattern style [fine pattern], brand [HLE-DY], popular elements [Europe and the United States], major downstream platforms [Amazon], and major sales areas [Africa].

[0086] The following types of keywords are recommended now, for example, the recent high-frequency search terms of buyers: [quilting], [lace], [printing], [coating], and competitors' selling points: [lace], [foam embroidery], [clothing], and [cotton thread]. You need to select 1 to 3 most appropriate keywords from the above product traffic words based on product information, add them to the original title, to help the product get more search traffic, and then output the recommended words and.

[0087] After receiving the information provided by the above agent, the large language model will give a response: "Based on the product information, the following keywords are recommended for you:

[0088]

Quiltment

[0089] [Lace]: This keyword is consistent with the specific purpose of the product (stage costumes) and can attract buyers who have demand for stage costumes.

[0090] [Panhua]: This keyword emphasizes the exquisite craftsmanship of the product and helps to enhance the attractiveness of the product.

[0091] After comprehensive consideration, it is recommended to add the three keywords [quilting], [lace] and [flowering] to the original title. This can better highlight the characteristics of the product, attract more potential buyers' attention, and thus obtain more search traffic. The final product title can be: [dyed quilting embroidery sequins three-dimensional leaves bead tube embroidery fabric cross-border children's clothing wedding dress mesh fabric mesh gauze-lace-flowering]. ”

[0092] In S107, a second candidate traffic word for adding to the original title is generated based on the product information and the first candidate traffic word using a large language model.

[0093] According to some embodiments, the third intelligent agent innovates in algorithm recommendation based on the product information and the first recommended traffic word, and uses a large language model to generate a second candidate traffic word for adding to the original title.

[0094] An example will be described below.

[0095] Input the following information to the large language model:

[0096] Now there is an e-commerce product with the following product information: the current title of the product is [Dyed Quilted Embroidery Sequins Three-dimensional Leaf Bead Tube Embroidery Fabric Cross-border Children's Clothing Wedding Dress Mesh Fabric Mesh]; the attributes of the product include the ingredients and content [Polyester], yarn count [as per sample], gram weight [0.92 (g / ㎡)], width [125-135], embroidery craft [sequin embroidery], origin [Guangdong], specific use [stage costume], applicable season [spring and summer], pattern style [fine pattern], brand [HLE-DY], popular elements [Europe and the United States], major downstream platforms [Amazon], and major sales areas [Africa].

[0097] The following types of keywords are recommended now. For example, the recent high-frequency search terms of buyers: [quilting embroidery], [lace], [printing], [floral], and competitors' selling points: [lace], [foam embroidery], [clothing], [cotton thread]. Based on the product information and recommended keywords, you need to innovate an additional 3 words. The length of the words should not exceed three characters, and give the reasons.

[0098] After receiving the information provided by the above intelligent agent, the large language model will give a response: "Based on the product information and recommended keywords, the following three keywords are created for the merchant:

[0099] [Three-dimensional]: Because the product description mentions "three-dimensional leaf bead tube embroidery fabric", which emphasizes the three-dimensional sense of the fabric, this keyword can highlight the uniqueness of the product.

[0100]

Light and thin

[0101] [Exquisite]: The pattern style of the product is described as "fine pattern". This word can be used to emphasize the delicacy and high quality of the product to attract consumers' attention.

[0102] Please note that the choice of keywords needs to be determined based on your target market and consumer group, and the above is only one possible direction. "

[0103] In S109, based on the product information and the first candidate traffic word and the second candidate traffic word, a large language model is used to determine the excluded risk traffic words to obtain a third candidate traffic word.

[0104] According to some embodiments, the fourth agent performs risk judgment using a large language model LLM and a risk control interface.

[0105] An example will be described below.

[0106] Input the following information to the large language model:

[0107] Now there is an e-commerce product with the following product information: the current title of the product is [Dyed Quilted Embroidery Sequins Three-dimensional Leaf Bead Tube Embroidery Fabric Cross-border Children's Clothing Wedding Dress Mesh Fabric Mesh]; the attributes of the product include the ingredients and content [Polyester], yarn count [as per sample], gram weight [0.92 (g / ㎡)], width [125-135], embroidery craft [sequin embroidery], origin [Guangdong], specific use [stage costume], applicable season [spring and summer], pattern style [fine pattern], brand [HLE-DY], popular elements [Europe and the United States], major downstream platforms [Amazon], and major sales areas [Africa].

[0108] The system recommends several optimization words:

[0109]

Quiltwork

[0110] [Sequin embroidery]: This keyword can emphasize the embroidery craftsmanship of the product, especially the high-end feel of sequin embroidery, and attract buyers who pursue unique designs.

[0111]

Stage Costumes

[0112] Intelligent recommendation (AI recommendation) has several optimized words:

[0113] [Three-dimensional]: Because the product description mentions "three-dimensional leaf bead tube embroidery fabric", which emphasizes the three-dimensional sense of the fabric, this keyword can highlight the uniqueness of the product.

[0114]

Light and thin

[0115] [Exquisite]: The pattern style of the product is described as "fine pattern". This word can be used to emphasize the delicacy and high quality of the product to attract consumers' attention.

[0116] For professional e-commerce industry operators, new keywords need to be added to product titles. On the premise that the titles are suitable for the products themselves, which of these recommended optimization words would be excluded from a professional perspective?

[0117] After receiving the information provided by the above agent, the large language model will give a response: "When adding keywords, it is necessary to consider the main features of the product and the target market. Based on the product information and the recommended optimization words, the following keywords will be excluded:

[0118] [Quilt embroidery]: Although this keyword can highlight the unique craftsmanship of the product, considering that the product title already contains the complete description of "Quilt embroidery sequins three-dimensional leaf bead tube embroidery fabric", repeatedly using "Quilt embroidery" may make the title appear lengthy and not concise enough.

[0119] [Thin and Light]: Although the product is light in weight, based on the product's purpose (stage costume) and main sales area (Africa), it may require thicker fabrics to provide better support and protection. Therefore, the word "thin and light" may not be the most appropriate description.

[0120] [Exquisite]: Although the pattern style of the product is described as "fine pattern", this word may remind people of a more refined and advanced design, which is not reflected in the other descriptions of the product. Therefore, the word "exquisite" may not be appropriate.

[0121] Based on the above analysis, the three keywords "quilting", "thin" and "exquisite" are excluded. "

[0122] In S111, based on the product information, an optimized traffic word is selected from the third candidate traffic word using a large language model and the optimized traffic word is added to the original title, thereby generating a new title.

[0123] According to the above process, we can see that the large language model screens the algorithm recommended optimization words, generates new optimization words, and screens again based on the results of the above two steps. Through the above steps, the optimized traffic words after risk judgment are added to the original product title to generate a new product title.

[0124] The method of the present invention can better help merchants screen traffic keywords and optimize product titles more conveniently. When many traffic words are generated but no priority is selected, it is impossible to determine which word is most suitable for the product, and these words may conflict with the attributes of the product itself, based on the actual situation of the product, a large language model can be combined to give recommended words and reasons for the recommendation.

[0125] According to some embodiments, the aforementioned method may be implemented by a plurality of agents associated with the algorithm recommendation system and the large language model (see below in conjunction with Figure 4 Detailed description). For example, the method of the present invention refers to an idea based on artificial intelligence and natural language processing technology (COT), designs a system for optimizing product titles, and guides the language model by designing multiple agents and using chain thinking prompts. The multiple agent modules are used for the service platform to send requests and return results, and the service platform includes a large language model and an algorithm recommendation system.

[0126] The plurality of agent modules may include first to fifth agents.

[0127] The first intelligent agent obtains product information, which includes the original title and product attributes of the product. The second intelligent agent uses an algorithm recommendation system to obtain relevant traffic words associated with the product, and uses a large language model to select a first candidate traffic word from the relevant traffic words to be added to the original title based on the product information. The third intelligent agent uses a large language model to generate a second candidate traffic word to be added to the original title based on the product information and the first recommended traffic word. The fourth intelligent agent uses a large language model to determine the excluded risk traffic words based on the product information, the first candidate traffic words, and the second candidate traffic words to obtain a third candidate traffic word. The fifth intelligent agent uses a large language model to select an optimized traffic word from the third candidate traffic word based on the product information and the large language model and adds the optimized traffic word to the original title, thereby generating a new title.

[0128] The multiple agent modules are implemented in the form of classes; the partition module combines at least a part of the multiple agent modules by dynamic instantiation. The pipeline processing flow module is implemented by the pipeline processing flow class and stores the context of the execution process of the multiple agent modules.

[0129] During the design process, it was found that the thinking process between adjacent agents is not a strict relationship of the previous output being the next input. Therefore, two parts are needed to realize multi-agent, one of which is the agent cluster (agent cluster) and the other is the pipeline process cluster (pipeline cluster). Through the agent cluster, the recommendation results are requested from the large language model and returned; through the pipeline process cluster, the agents are selected, sorted and combined, and the dialogue information during the execution of the large language model is stored.

[0130] According to some embodiments, after obtaining the filtered commodity traffic words, the original commodity title is optimized according to the actual situation, such as adding new commodity traffic words, deleting old commodity traffic words that reduce exposure, etc.

[0131] According to the method of the present invention, merchants optimized the number of products 10 times on average. After optimization, the merchants and products hit by optimized word searches accounted for 69% and 33% respectively. The positive exposure of optimized products accounted for 59.12%, of which 58.63% increased significantly.

[0132] Here is an example of how to optimize product titles. A conversation is initiated with a large language model, and the conversation agent content is as follows:

[0133] Now there is an e-commerce product with the following product information:

[0134] The current title of the product is: Cartoon Household Fruit Peeler Multifunctional Kitchen Planer Portable Peeling 2 in 1;

[0135] The types of goods are: daily dining and drinking utensils / kitchen tools / planes and peelers;

[0136] The attributes of the product are: Brand: [KUCOCO]; Whether imported: [No]; Place of origin: [Yiwu]; Product launch time: [Spring 2022]; Item number: [YBJ-26500]; Custom processing: [Yes]; LOGO printing: [Yes]; Price range: [5-10 yuan]; Whether patented: [No]; Suitable gift-giving occasions: [Birthday]; Material: [ABS+PC+stainless steel 201].

[0137] According to the existing system algorithm recommended words, the optimized words and their selection reasons are as follows:

[0138] Scraping: It is consistent with the product positioning and is a high-frequency search term;

[0139] Multi-purpose: consistent with product functions and high-frequency search terms;

[0140] Fruits and vegetables: They are consistent with the product’s usage scope and are high-frequency search terms;

[0141] Storage: It is consistent with the additional functions of the product and does not conflict with the title;

[0142] Pumpkin: Keywords that match the target users of the product and are selling points of competitors.

[0143] Based on the existing product information, the optimization words generated by AI and the reasons for their selection are as follows:

[0144] Scrape: Improve exposure because it is a practical functional description related to peeling, which can attract the attention of target consumers;

[0145] Portable: Increase exposure because it highlights the convenience of the product and attracts more users’ interest;

[0146] Chipping: Improve exposure, because it is a detailed description of product features, increasing search relevance;

[0147] Kitchen: Improve exposure because it clarifies the usage scenario and helps users quickly locate the desired products;

[0148] Colorful: Increase exposure because it describes the product's appearance and attracts more visually sensitive consumers.

[0149] Combined with the product information, the above-screened optimization words are excluded. The excluded keywords and the reasons for their exclusion are as follows:

[0150] Scraping and slicing: This does not conform to the actual functional description of the product. The product is mainly used for peeling fruits and vegetables, not slicing and slicing;

[0151] Chipping: It is repeated with the "peeling" function in the existing title, so there is no need to emphasize it again;

[0152] Kitchen: Repeats the "kitchen" scene in the existing title and does not need to be emphasized again;

[0153] Colorful: The title and attributes do not mention the variety of product colors, so it is impossible to confirm whether this is true;

[0154] Please note that the above exclusion reasons have taken into account the principles of not conflicting with product attributes and avoiding large amounts of duplication with title terms.

[0155] For professional e-commerce operations, if you want to add new keywords to the product title without adding other content to increase traffic, please select an optimization word and add it to the appropriate position of the title without adding any other content. The input information includes four parts: basic product information, optimization words recommended by the system algorithm, optimization words generated by AI, and optimization words excluded from the above two types of optimization words. Please select the preferred optimization words according to the following requirements:

[0156] Combine the optimization words recommended by the system algorithm and the optimization words generated by AI into a candidate set; remove the excluded optimization words from the candidate set; select a preferred optimization word based on the situation of the product itself, and add it to the appropriate position of the title to generate a new title. Please be careful not to add any content other than the selected preferred optimization word.

[0157] Please think according to the above process but do not output the thinking process. The output is two parts: the optimization word finally selected and the reason why the optimization word can improve exposure. The reason is summarized in one sentence. The two are separated by "," and arranged in the above order. Please output in plain text without any format including slogan language (markdown).

[0158] According to the process, we can see that the big model screened the optimization words recommended by the algorithm, generated new optimization words, and screened again based on the results of the above two steps. Although the big model has been very effective in text processing at this stage, there is still a possibility of error. For example, the optimization words generated by the above AI appear to be repeated or conflicting with the product title, especially when the thinking process is ignored during output. Setting up a thinking framework through multi-agent processing can effectively reduce this situation. At the same time, when the algorithm does not verify the optimization words and product attributes, and thus gives a result that is inconsistent with the product attributes, it can also be filtered out through the big model.

[0159] newWord (new keyword) = multi-purpose, newWordType (new keyword type) = m (quantifier), optimizeReason (optimization reason) = consistent with product functions and is a high-frequency search term.

[0160] Figure 2 A schematic diagram illustrating a detailed optimization process according to an example embodiment.

[0161] See also Figure 2 ,According to the custom word segmentation dictionary, the original title is tagged with custom word segmentation parts of speech, and reserved words such as brand, model, etc. are determined, and low-exposure words and optional words such as the word "factory direct sales" are determined.

[0162] If the number of characters in the entire title after the reserved words is greater than 54, or there are low-exposure words or optional words, then the word reduction operation is performed; if the number of characters in the entire title after the reserved words is less than 54, the query words, drainage words, selling point words and words with high word frequency are weighted for similarity calculation, and the large language model is used to select product traffic words based on the correlation measurement, and then the word addition operation is performed. Combined with the word addition and subtraction operations, a new title is generated. When the new title is changed seven days later, the product exposure rate is determined. If the exposure rate decreases by 20%, the title is modified again.

[0163] The following is an example of filtering product traffic words.

[0164]

[0165] Cloth mesh",

[0166] "newWord":"Lace",

[0167] },

[0168] {

[0169] "newTitle":"Dyed embroidery sequins three-dimensional foam embroidered leaves bead tube fabric cross-border children's clothing wedding dress mesh fabric mesh",

[0170] "newWord":"Foam embroidery",

[0171] },

[0172] {

[0173] "newTitle":"Dyed embroidered sequins three-dimensional leaves beaded tube fabric cross-border children's clothing wedding dress dress mesh fabric mesh",

[0174] "newWord":"clothing",

[0175] },

[0176] {

[0177] "newTitle":"Dyed embroidery sequins cotton thread three-dimensional leaf bead tube fabric cross-border children's clothing wedding dress mesh fabric mesh",

[0178] "newWord":"cotton thread",

[0179] }

[0180] ],

[0181] }

[0182] Figure 3 A schematic diagram showing a system for optimizing product titles according to an example embodiment.

[0183] See also Figure 3 The system for optimizing product titles of the present invention includes: an intelligent body module, which is used to send requests to a large language model and return results; a partitioning module, which is used to combine intelligent body modules; and a pipeline process processing module, which is responsible for arranging the execution order of the partitioning modules and setting checkpoints.

[0184] Figure 4 A schematic diagram showing a multi-agent implementation according to an example embodiment.

[0185] The component modules of the system / device are implemented as classes; the agent module, the partition module and the pipeline process processing module are allocated and scheduled through dynamic assembly. The agent module is implemented by the agent class, which requests and returns recommendation results from the large language model; the partition module is implemented by the partition class, which is used to combine agent classes; the pipeline processing flow module is implemented by the pipeline processing flow class, which is responsible for arranging the execution order of the partition module and setting checkpoints, and storing the dialogue information during the execution of the large language model.

[0186] The present invention adopts a chain thinking method. Chain thinking is structured thinking, which provides a systematic way of thinking, making the discussion and decision-making process more organized and avoiding confusion in thinking. Chain thinking can consider problems from all angles. By thinking about problems from different angles (facts, emotions, optimism, pessimism, creativity, control), it can ensure that the problems are fully analyzed and avoid one-sidedness. Chain thinking can improve the quality of decision-making, because each thinking node has a clear thinking direction, which can help the team to analyze more deeply and comprehensively when making decisions, thereby reaching higher quality decisions. Chain thinking can reduce emotional interference. By separating emotions from facts, it can effectively reduce the negative impact of emotions on the decision-making process, making the discussion more objective and fair.

[0187] The method of the present invention can combine agent classes by section, so as to cope with some situations with complex application logic, which belongs to the category of chain call as a whole. If it is just a simple process, it can be simply arranged and executed in the above way. If the execution nodes are more than 20, it is better to use a more complex process arrangement framework to implement it. If you encounter more complex situations that are difficult to deal with in the future, you can introduce some heavyweight frameworks to solve complex problems.

[0188] Figure 5 A schematic diagram showing a multi-agent design according to an example embodiment.

[0189] The first agent of the multi-agent system focuses on objective facts and data and analyzes existing information; the second agent of the multi-agent system symbolizes optimism and looks for positive aspects and opportunities in problems; the third agent of the multi-agent system represents creative thinking and encourages new ideas and solutions; the fourth agent of the multi-agent system is used for critical thinking and identifies potential problems and risks; the fifth agent of the multi-agent system is responsible for organizing and controlling the thinking process to ensure that thinking is carried out in an orderly manner.

[0190] During the design process, it was found that the relationship between adjacent agents is not strictly the previous output is the next input. For example, the inputs of the second and third agents are the same, and their outputs become part of the fourth agent. In addition, the fourth and fifth agents will also use the outputs of non-previous nodes, so this is not a strict chain structure. The specific execution process is as follows: Figure 3 shown.

[0191] Therefore, two parts are needed to implement multi-agent. One part is the agent class cluster, which is responsible for requesting and returning results from the large model. The other part is the pipeline class cluster, which is responsible for selecting, sorting or combining agents and storing the context of the execution process, such as Figure 4 shown.

[0192] The agent class is responsible for requesting and returning results from the large model. Its main execution methods are:

[0193] public Object execute(Object params,HashMap<String,Object> context){

[0194] params=preExecute(params,context);

[0195] Object result=request(params);

[0196] return postExecute(result,context);

[0197] }

[0198] Among them, the preExecute method is used to process parameters to meet the requirements of the request to the big model; the request method is to specifically request the big model and return the result; the postExecute method is used for post-processing, including processing the result and storing it in the context. These methods can be overridden by subclasses to achieve specific needs. The agent class needs to be combined by the section class. The main execution methods of the section class are:

[0199] public Object execute(Object params,HashMap<String,Objec t> context){

[0200] if(CollectionUtil.isEmpty(agents)){

[0201] return params;

[0202] }

[0203] return agentExecute(params);

[0204] }

[0205] The role of agentExecute is to execute all agents in the section. The default execution is sequential, which can be modified to concurrent execution.

[0206] The pipeline class is responsible for arranging the section execution order and setting checkpoints. If a section is set as a checkpoint, its return result will be checked. If it is null, the entire process ends. The main execution methods of the pipeline class are:

[0207] public Object execute(Object params){

[0208] AgentSection now=head;

[0209] while(now!=null){

[0210] params=now.execute(params,context);

[0211] / / Checkpoint judgment

[0212] if(checkPoint.contains(now.getName())&&(params==null))

[0213] {

[0214] return null;

[0215] }

[0216] now = now.next();

[0217] }

[0218] return params;

[0219] }

[0220] When finally called, initialize the section and fill it with the corresponding agent, and then sort the section through the pipeline.

[0221] The implementation of multi-agent is as follows:

[0222] / / First Agent

[0223] AgentSection whiteHat=new WhiteHatSection();

[0224] whiteHat.setAgents(Lists.newArrayList(AgentFactory.getAgent(CopilotAutomati cAgentEnum.TitleBaseInfoAgent.getCode())));

[0225] / / The second and third agents

[0226] AgentSection yellowAndGreenHat=new YellowAndGreenHatSecti on();

[0227] yellowAndGreenHat.setAgents(Lists.newArrayList(AgentFactory.getAgent(CopilotAutomaticAgentEnum.TitleLLMKeywordsAgent.getCode()), AgentFactory.getAgent(CopilotAutomaticAgentEnum.TitleSysKeywordsAgent.getCode())));

[0228] / / Fourth intelligent agent

[0229] AgentSection blackHat = new BlackHatSection();

[0230] blackHat.setAgents(Lists.newArrayList(AgentFactory.getAgent(CopilotAutomaticAgentEnum.TitleExcludeAgent.getCode())));

[0231] / / Fifth intelligent agent

[0232] AgentSection blueHat = new BlueHatSection();

[0233] blueHat.setAgents(Lists.newArrayList(AgentFactory.getAgent(CopilotAutomaticAgentEnum.TitleSummaryAgent.getCode())));

[0234] / / Initialize the execution process

[0235] AgentPipeLine thinkingHats = new AgentPipeLine();

[0236] thinkingHats.add(whiteHat, true).add(yellowAndGreenHat).add(blackHat).add(blueHat);

[0237] / / Start execution

[0238] thinkingHats.execute(params)

[0239] If you want to select only one of the second agent or the third agent, you can set it when filling in the agent. If you want to use the multi-agent framework in other tasks, you can rewrite the agent and fill it in the corresponding section.

[0240] Although the large language model in the prior art has been very effective in text processing, it is still possible to make mistakes, such as the generated optimization words appearing in the content that is repeated or conflicting with the product title, especially when the thinking process is ignored during output. The method of the present invention can effectively reduce this situation by setting up a thinking framework for processing through multiple agents. At the same time, when the algorithm does not verify the optimization words and product attributes and gives a result that is inconsistent with the product attributes, it can also be filtered out through the large model.

[0241] Figure 6 A block diagram of a computing device is shown according to an example embodiment.

[0242] like Figure 6 As shown, computing device 30 includes processor 12 and memory 14. Computing device 30 may also include bus 22, network interface 16, and I / O interface 18. Processor 12, memory 14, network interface 16, and I / O interface 18 may communicate with each other via bus 22.

[0243] The processor 12 may include one or more general-purpose CPUs (Central Processing Units, processors), microprocessors, or application-specific integrated circuits, etc., for executing relevant program instructions. According to some embodiments, the computing device 30 may also include a high-performance graphics card (GPU) 20 for accelerating the processor 12.

[0244] The memory 14 may include a machine system readable medium in the form of a volatile memory, such as a random access memory (RAM), a read-only memory (ROM) and / or a cache memory. The memory 14 is used to store one or more programs including instructions and data. The processor 12 can read the instructions stored in the memory 14 to execute the above-mentioned method according to the embodiment of the present invention.

[0245] The computing device 30 may also communicate with one or more networks via the network interface 16. The network interface 16 may be a wireless network interface.

[0246] The bus 22 may include an address bus, a data bus, a control bus, etc. The bus 22 provides a path for exchanging information between components.

[0247] It should be noted that, in the specific implementation process, the computing device 30 may also include other components necessary for normal operation. In addition, those skilled in the art may understand that the above device may only include components necessary for implementing the embodiments of this specification, and need not include all components shown in the figure.

[0248] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above method when executed by a processor. The computer-readable storage medium may include, but is not limited to, any type of disk, including a floppy disk, an optical disk, a DVD, a CD-ROM, a microdrive, and a magneto-optical disk, a ROM, a RAM, an EPROM, an EEPROM, a DRAM, a VRAM, a flash memory device, a magnetic card or an optical card, a nanosystem (including a molecular memory IC), a network storage device, a cloud storage device, or any type of medium or device suitable for storing instructions and / or data.

[0249] An embodiment of the present invention also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute part or all of the steps of any method recorded in the above method embodiments.

[0250] Those skilled in the art can clearly understand that the technical solution of the present invention can be implemented with the help of software and / or hardware. "Unit" and "module" in this specification refer to software and / or hardware that can independently complete or cooperate with other components to complete specific functions, where the hardware can be, for example, a field programmable gate array, an integrated circuit, etc.

[0251] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0252] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0253] In the several embodiments provided by the present invention, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0254] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0255] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0256] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product, which is stored in a memory and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the methods of various embodiments of the present invention.

[0257] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0258] The exemplary embodiments of the present invention are specifically shown and described above. It should be understood that the present invention is not limited to the detailed structure, configuration or implementation method described herein; on the contrary, the present invention is intended to cover various modifications and equivalent configurations included in the spirit and scope of the attached clauses.

Claims

1. A method for optimizing a product title, comprising: Obtaining product information, including the original title and product attributes of the product; Using an algorithm recommendation system to obtain relevant traffic words associated with the product; Selecting a first candidate traffic word for adding to the original title from the relevant traffic words using a large language model based on the product information; Generate a second candidate traffic word for adding to the original title using a large language model based on the product information and the first candidate traffic word; Based on the product information and the first candidate traffic word and the second candidate traffic word, a large language model is used to determine and exclude risky traffic words to obtain a third candidate traffic word; Based on the product information, a large language model is used to select an optimized traffic word from the third candidate traffic word and the optimized traffic word is added to the original title, thereby generating a new title.

2. The method according to claim 1, characterized in that The relevant traffic words include: The query keywords with high exposure for the products in the corresponding leaf categories within the predetermined time period; Selling point keywords extracted from the titles of competing products; The product’s own traffic-generating keywords.

3. The method according to claim 1, characterized in that Based on the recommendation algorithm, relevant traffic words associated with the product are obtained, including: Performing a weighted semantic similarity calculation on the commodity traffic word, the category name of the commodity and the original title of the commodity; According to the weighted calculation result, the relevant traffic words are obtained using a pre-trained small model.

4. The method according to claim 1, characterized in that: Also includes: Through natural language processing tools, Chinese words with a predetermined frequency are selected from the word segmentation of all titles; Build a maintainable custom word segmentation dictionary using the Chinese vocabulary; The original title is split using the custom word segmentation dictionary.

5. The method according to claim 1, characterized in that The method is implemented by a plurality of agents associated with the algorithm recommendation system and the large language model.

6. A system for optimizing product titles, the system comprising: A plurality of intelligent agent modules, wherein the plurality of intelligent agent modules are used for a service platform to send requests and return results, wherein the service platform includes a large language model and an algorithm recommendation system; a partitioning module, the partitioning module being used to combine at least a portion of the plurality of agent modules; A pipeline process processing module, the pipeline process processing module is used to set the execution order of the partition modules and set result checkpoints, Wherein, the multiple agent modules include: Case Number: The first agent obtains product information, where the product information includes the original title and product attributes of the product; The second agent uses an algorithm recommendation system to obtain relevant traffic words associated with the product, and selects a first candidate traffic word for adding to the original title from the relevant traffic words based on the product information using a large language model; A third agent generates a second candidate traffic word for adding to the original title based on the product information and the first recommended traffic word using a large language model; A fourth intelligent agent determines and excludes risky traffic words based on the commodity information, the first candidate traffic word, and the second candidate traffic word using a large language model to obtain a third candidate traffic word; The fifth intelligent agent uses a large language model to select optimized traffic words from the third candidate traffic words based on the product information and adds the optimized traffic words to the original title, thereby generating a new title.

7. The system according to claim 6, characterized in that The multiple agent modules are implemented in a class manner; The partition module realizes combining at least a part of the multiple agent modules by means of dynamic instantiation.

8. The system according to claim 7, characterized in that The pipeline processing flow module is implemented by a pipeline processing flow class and stores the context of the execution process of the multiple intelligent agent modules.

9. A computer program product, characterized in that The method comprises a computer program, which implements the method according to any one of claims 1 to 5 when being executed by a processor.

10. A computing device, characterized in that include: processor; as well as A memory storing a computer program, which, when executed by the processor, enables the processor to perform the method according to any one of claims 1 to 5.

Images