1688平台算法原理及数据处理情况说明
Explanations of Algorithms and Data Processing on 1688 Platform
为依法保障用户对算法推荐服务的基本原理、目的意图和主要运行机制的知情权和选择权，告知用户1688平台提供的算法推荐服务基本情况，1688平台服务提供者（或简称“我们”）特制定本《1688平台算法原理及数据处理情况说明》，帮助用户充分了解在使用1688平台产品或服务的过程中，了解我们如何通过利用个性化推送类、检索过滤类、排序精选类算法技术向用户提供信息或服务，充分保障用户合法权益。
For the purposes of protecting users’ right to know the basics, purposes and operating mechanism of algorithmic recommendation services of 1688 Platform and their right of choice under laws, these Explanations of Algorithms and Data Processing on 1688 Platform (these “Explanations”) are hereby formulated by 1688 platform service providers (“we”, “us” or “our”) to help users know what algorithmic recommendation services we provide on 1688 Platform and fully understand how we provide them with information or services by algorithmic technologies, such as those for personalized push notifications, retrieval and filtering, or ranking and selection, in their use of products or services on 1688 Platform, so as to fully protect their legitimate rights and interests.
一、适用范围
I. Scope of Application
本说明适用于1688平台以网站、客户端、小程序等形式，向您提供的各项产品或服务。
These Explanations shall apply to all products or services provided by 1688 Platform to you through its websites, applications, applets, etc.
二、算法原理说明
II. Explanations of Algorithms
1、个性化推送类算法
Algorithms for Personalized Push Notifications
算法名称 Name | 1688个性化推送类算法 Algorithms for Personalized Push Notifications of 1688 Platform |
算法基本原理 Basics | 为向1688平台电商用户展示商品或服务信息，包括用户的访问足迹、历史搜索情况，我们会收集和使用用户在访问或使用1688时的浏览、搜索记录。我们会结合依法收集的设备信息、服务日志信息，以及其他取得用户授权的信息，通过算法模型预测人群的偏好特征。我们会基于人群的偏好特征在1688应用程序向相关人群推送用户可能感兴趣的商业广告及其他信息。 We will collect and use the browsing and search records of e-commerce users on 1688 Platform generated in their visit or use of 1688 Platform, in order to show them information on commodities or services, including their browsing and search histories. Combined with the legally collected device information, service log information and such other information as authorized by users, we will predict the preferences of each user group by algorithmic models, and push to each user group commercial advertisements and other information in which they may be interested through 1688 applications. |
算法运行机制 Operating Mechanism | 个性化推送算法会基于模型预测人群的偏好特征，匹配人群可能感兴趣的商品、服务或其他信息，对向用户展示的商品、服务或其他信息进行排序。我们会根据用户使用产品过程中的浏览行为，对推荐模型进行实时反馈，不断调整优化推荐结果。为满足用户的多元需求，我们会在排序过程中引入多样化推荐技术，拓展推荐的内容，避免同类型内容过度集中。 The algorithms for personalized push notifications will, based on models, predict the preferences of each user group, match with them the commodities, services or other information in which they may be interested, and rank the commodities, services or other information to be displayed to them. We will keep optimizing the recommendation results based on the real-time feedback of browsing behavior of users to the recommendation models in their use of products. In order to satisfy the diversified needs of users, we will introduce diversified recommendation technologies into the ranking process, thus expanding the scope of recommended contents, and avoiding excessive concentration of the same type of contents. 如用户不想看到我们在首页或进入订单页面给用户推荐的商品或服务，用户可以通过长按被推荐的商品或服务图片，在随后出现的弹窗中根据提示选择屏蔽类似商品或者商品或服务所属的类目；如用户想管理我们给用户推送的个性化内容，用户可以在“我的-设置-隐私-推荐管理-个性化内容推荐设置”中进行设置。 If a user does not want to see any commodity or service recommended by us to him/her on the homepage or order page, he/she may long press the image of such recommended commodity or service, and follow the instructions on the pop-up window to block either similar commodities or a category of commodities or services. If a user intends to manage the personalized contents pushed by us to him/her, he/she may do so through “My - Settings - Privacy - Recommendation Management - Settings of Personalized Recommendations.” |
算法应用场景 Application Scenarios | 1688平台（阿里巴巴（1688）APP、1688工业品APP、1688网站）首页支付完成页面等的商品或服务信息展示 Homepage, successful payment page and other pages of 1688 Platform (including 1688 application, 1688 industry application and 1688.com) |
算法目的意图 Purpose | 向用户展示商品或服务信息 To display information on commodities or services to users |
备案编号 Filing No. |
2、内容安全算法
Algorithms for Content Safety
算法名称 Name | 1688内容安全算法 Algorithms for Content Safety of 1688 Platform |
算法基本原理 Basics | 我们基于大量样本数据的分析，形成内容安全算法模型，依法对1688平台上发布的文本、图片、音频、视频等信息内容进行识别和处置，防范违反相关法律法规规定的淫秽、色情、赌博、暴力、恐怖、教唆犯罪、欺诈、虚假、侮辱、诽谤、恐吓、封建迷信等信息，以及可能侵害他人隐私、知识产权等合法权益的信息的发布和传播。 We have created algorithmic models for content safety based on the analysis of a large size of sample data, to identify and handle the texts, images, audios, videos and other information that are published on 1688 Platform in accordance with laws, in order to prevent the publishing and spread of information on, amongst others, obscenity, pornography, gambling, violence, terror, instigation of crime, fraud, falsehood, insult, defamation, intimidation and feudal superstition that is in violation of applicable laws and regulations, and information that may infringe upon the privacy, intellectual property rights or other legitimate rights and interests of others. |
算法运行机制 Operating Mechanism | 内容安全算法的运行过程包括数据源接入、算法识别、审核、处置决策等。我们对1688平台上文本、图片、音频、视频等信息，通过深度学习、知识图谱推理、时序模型和融合模型等风险分类模型进行安全风险识别，形成不同的风险置信度分级，对于高置信度的信息由算法直接完成审核，对于低置信度的信息将引入人工审核，最后完成对违规信息的处置决策。 The operating process of the algorithms for content safety includes, amongst others, data source access, algorithmic identification, review and decision-making. We identify safety risks in texts, images, audios, videos and other information on 1688 Platform through risk classification models, such as deep learning, knowledge graph inference, time series model and integration model, so as to classify such information into different levels of risk confidence. Information with a high confidence level will be reviewed by algorithms directly, while that with a low confidence level will be subject to manual review, following which a final decision will be made in respect of non-compliant information. |
算法应用场景 Application Scenarios | 阿里巴巴（1688）APP和1688网站上的商品、评价、问答、论坛等信息发布相关的所有场景 All scenarios relating to the publishing of information on 1688 application and 1688.com, such as commodities, comments, Q&A and forums 1688工业品APP上的商品、评价、问答等信息发布相关的所有场景 All scenarios relating to the publishing of information on 1688 industry application, such as commodities, comments and Q&A |
算法目的意图 Purpose | 及时发现、处置违反法律、行政法规或违反社会公德、公序良俗的信息内容。 To promptly identify and handle information that is in violation of laws, administrative regulations, social morality, public order or social customs |
备案编号 Filing No. |
3、检索类算法
Algorithms for Retrieval
算法名称 Name | 1688检索类算法 Algorithms for Retrieval of 1688 Platform |
算法基本原理 Basics | 为向1688平台用户展示更契合检索意图的商品或服务信息，检索类算法将针对用户输入的搜索词，使用算法模型预测、匹配其相关可能感兴趣的商品或服务信息，最终完成检索结果的展示。 To provide the users of 1688 Platform with information on commodities or services that better meet their purposes of retrieval, we use algorithmic models to predict and match information on commodities or services in which the users may be interested, based on the keywords input by them, and display the final results of retrieval. |
算法运行机制 Operating Mechanism | 在用户输入的搜索词后，我们将使用文本匹配的倒排索引和基于神经网络的向量匹配召回算法，根据搜索关键词特征、商品特征、用户在1688的使用情况等对商品和服务进行召回，并结合相关性模型保障结果页展现的商品结果与搜索的关键词相关。同时，为满足用户多元需求，我们会在排序过程中引入多样性打散机制，拓展展示的内容，避免同类型内容过度集中。 After a user inputs a keyword, we will use the inverted indexing based on text matching and the vector fitting & matching algorithm based on neural network to match commodities and services based on, amongst others, the characteristics of keyword, characteristics of commodities and user behavior, and will guarantee that the commodities displayed as results on the result pages will be relevant to the keyword based on relevance models. In addition, in order to satisfy the diversified needs of users, we will introduce a mechanism of shuffling into the ranking process, thus expanding the scope of displayed contents, and avoiding excessive concentration of the same type of contents. 用户在使用我们提供的站内搜索服务时，需要查看不针对其个人特征的排序，可以在如搜索结果页面点击“筛选”，选择其中的“销量”“价格”“通用排序”进行设置等。 If a user needs to view the search results in an order that is not specific to his/her personal characteristics when using the retrieval service provided by us on 1688 Platform, he/she may click “filter” on the search result page, and sort the results by choosing, amongst others, “sales volume”, “price” or “general”. |
算法应用场景 Application Scenarios | 1688平台（阿里巴巴（1688）APP、1688工业品APP、1688网站）首页搜索框等商品或服务的检索 Retrieval of commodities or services through the search box, amongst others, on the homepage of 1688 Platform (including 1688 application, 1688 industry application and 1688.com) |
算法目的意图 Purpose | 帮助用户快速找到想要的商品或信息 To help users quickly find the commodity or information they need |
备案编号 Filing No. |
4、生成合成类算法
Algorithms for Generation and Synthesis
算法名称 Name | 1688 阿牛智能客服算法 A’niu Smart Customer Service Algorithm of 1688 Platform |
算法基本原理 Basics | 1688 平台根据用户咨询内容，结合阿牛智能客服知识库，利用自然语言处理技术定位用户需要的知识，并给出对应的解决方 案;在用户获得解决方案之后，利用对历史咨询数据的统计分析预 估下一阶段可能咨询的问题，帮助用户更快速、便捷地解决问题。 According to the inquiries from users, 1688 Platform will locate the knowledge required by users, and provide corresponding solutions using natural language processing technologies and A’niu smart customer services repository. Afterwards, 1688 Platform will conduct statistical analysis of the historical inquiry data of users to predict their future possible inquiries, so as to help them solve problems more quickly and conveniently. |
算法运行机制 Operating Mechanism | 用户通过客服进行了相关咨询，咨询的内容，经过去标识化 处理，在无法识别用户身份的情况下，1688 平台以问题为维度进 行抽样，用于智能客服算法模型训练，用于不断提升用户体验。 在用户接受智能客服服务期间，我们会对用户进行显著提示，基 于智能客服使用情况，不断改进通知客服的服务质量。 Inquiries from users will be de-identified and sampled by 1688 Platform for training the smart customer service algorithm models, in order to continuously improve user experience. When users use the smart customer service, we will give them prominent prompts, and based on their use of such service, we will continuously improve the smart customer service quality. |
算法应用场景 Application Scenarios | 1688APP、1688工业品APP、1688PC页面的阿牛智能客服 A’niu smart customer service in 1688 application and 1688 industry application and on 1688.com |
算法目的意图 Purpose | 较人工客服更高效、快捷地响应用户咨询 To respond to user inquiries more efficiently and conveniently |
备案编号 Filing No. |
5、深度合成服务
Deep Synthesis Services
算法名称 Name | 1688图像生成算法 Image Generation Algorithms of 1688 Platform |
算法基本原理 Basics | 我们使用图像生成算法，通过交互式对话方式，为用户提供图片创作工具。1688图像生成算法使用了基于自然语言处理技术的对话生成模型来分析用户输入的prompt数据，以及图文匹配模型来引导图像扩散过程。用户需输入文本数据、图像数据，勾选期待生成的模版图像，经模型处理后，算法可最终向用户输出融合用户期待效果与用户创作指令的合成图像。 We utilize image generation algorithms to provide users with image creation tools through an interactive conversational approach. The Image Generation Algorithms of 1688 Platform employs a dialogue generation model based on natural language processing technology to analyze user inputs from prompts, as well as an image-text matching model to guide the image diffusion process. Users are required to input text data and image data, select the desired template images, and after processing by the model, the algorithm can ultimately output a composite image that blends the expected effects with the creative instructions of the user. |
算法运行机制 Operating Mechanism | 1688图像生成算法属于结合图文匹配神经网络模型和扩散生成模型的生成合成类算法。在用户输入内容后，算法对用户的原始输入数据做预处理，将数据送往位于图形处理器的模型上，对数据进行运算，推理生成字符串格式的生成图片。整个扩散过程中，图文匹配模型基于扩散中间图和内容一致的图像。在此过程中，我们会通过安全过滤模型分别对输入、输入-输出进行风险检测。 Image Generation Algorithms of 1688 Platform belongs to the generative synthesis class of algorithms combining the graphic matching neural network model and the diffusion generation model. After the user inputs the content, the algorithm preprocesses the user's original input data, sends the data to the model located in the graphics processor, operates on the data, and reasons to generate the generated image in string format. Throughout the diffusion process, the graphic matching model guides the diffusion process based on the similarity between the diffusion intermediate graph and the input content, causing the diffusion model to generate images that are consistent with the content. In this process, we will detect the risk of the input and input-output through the security filtering model respectively. |
算法应用场景 Application Scenarios | 1688平台、1688APP上涉及图片创作的功能，如虚拟试衣 Functions involving image creation on the 1688 platform and 1688 APP, such as virtual fitting |
算法目的意图 Purpose | 通过智能对话，基于用户上传的图片生成体现用户创作意图的图片 Generate images that reflect the user's creative intent based on the images uploaded by the user through intelligent dialogs |
备案编号 Filing No. |
算法名称 Name | 1688对话生成算法 Dialog Generation Algorithms of 1688 Platform |
算法基本原理 Basics | 1688对话生成算法通过交互式对话方式，为用户提供购物搜索、信息内容服务等，最终提升电商平台购物体验。对话生成算法使用了基于自然语言处理技术的对话生成模型，它通过大量的数据训练模型来模拟人类的语言交互能力，实现在各种对话场景下的聊天对话。算法在线应用时，会使用用户实时输入的文本数据，在安全过滤的基础上，结合对话场景等向用户生成准确、得体的文本回复，部分场景下还会展示商品图片信息。 The Dialog Generation Algorithms of 1688 Platform provides users with shopping search, information content services, etc. through interactive dialog, which ultimately improves the shopping experience on the e-commerce platform. The dialogue generation algorithm uses a dialogue generation model based on natural language processing technology, which simulates the human language interaction ability through a large amount of data training model to realize the chat conversation in a variety of dialogue scenarios. When the algorithm is applied online, it will use real-time text data inputted by users, and on the basis of security filtering, it will generate accurate and decent text replies to the users in combination with the dialog scenes, and in some scenarios, it will also display product picture information. |
算法运行机制 Operating Mechanism | 1688对话生成算法使用Transformer神经网络架构，以预训练和微调技术为核心。在用户输入问题后，将依据意图分类模型识别出意图类型，生成符合用户消费习惯、搜索目的的文本。1688对话生成算法的训练数据来源于1688平台合法取得的数据。在此过程中，我们会通过安全过滤模型分别对输入、输入-输出进行风险检测。如我们发现用户输入的内容违反法律、行政法规等有关规定，我们将依法及时采取拦截、消除等处置措施。此类服务在用户选择特定功能或服务后启用，如用户不需要此类服务，可通过我们提供的指引自行关闭。 The Dialog Generation Algorithms of 1688 Platform uses the Transformer neural network architecture, with pre-training and fine-tuning techniques at its core. After the user inputs a question, the intent type will be identified based on the intent classification model, and the text that matches the user's consumption habits and search purpose will be generated.The training data of 1688 Dialogue Generation Algorithms comes from the legally obtained data on 1688 platform. In this process, we will detect the risk of input and input-output through the security filtering model. If we find that user input violates laws, administrative regulations and other relevant provisions, we will take timely interception, elimination and other disposal measures in accordance with the law. Such services are enabled after the user selects a specific function or service, and if the user does not need such services, he/she can turn them off by himself/herself through the guidelines provided by us. |
算法应用场景 Application Scenarios | 1688平台首页、客服页面、商品页面等的商品或信息内容展示，如智能导购、智能客服、智能问答等。 1688 platform home page, customer service page, product page and other products or information content display, such as intelligent shopping guide, intelligent customer service, intelligent Q&A. |
算法目的意图 Purpose | 向用户展示商品信息、提供信息内容服务，提高电商平台信息服务的趣味性。 Display product information to users, provide information content services, and improve the interest of information services on e-commerce platforms |
备案编号 Filing No. |