“忽如一夜春风来，千树万树梨花开”这句妙曼的诗句用来形容当下酒店业行业大模型的涌现十分贴切。10月27日，“2023酒店住宿餐饮业数字化发展峰会”举行的联合发布会上，云迹、辉驿、众荟、首旅如家、迈点联袂发布了各自的“酒店AI大模型”，一时真有点让人目不暇接。 数字化转型网www.szhzxw.cn
这是中国酒店业与新潮数字科技跟得最紧的一次。
去年末，OpenAI的ChatGPT横空出世，掀起一股人工智能的狂潮，人们被AGI大模型的绝世本领惊得“花容失色”。今年3月初杭州特勤局邀请我为西湖国宾馆高管讲课，特地要求增加ChatGPT内容。我用ChatGPT3.0自动书写了一套介绍ChatGPT来龙去脉的PPT，并对GPT的优劣进行了一次实例讲解。说实话此时ChatGPT虽有一本正经说瞎话的毛病，但其整体实力深深震惊了我，我意识到酒店行业一旦用上了这种所谓“大模型”，无疑会使数字化进程如虎添翼。
在随后短短的半年时间里，国内通用大模型高速发展，如雨后春笋。百度“文心一言”、阿里的“通义千问”、华为的“盘古”、科大讯飞的“星火”、中国移动的“九天众擎”、中国联通“鸿湖”等等几十家大模型先后诞生。虽然性能上，与ChatGPT4.0相比有些距离，但在总体功能上也很管用，假以时日和经过大量的中文材料训练，在质量上赶上国外产品颇有希望。
ChatGPT和大量国内通用大模型的出现，以其开源的特性和溢出效应将其所具有的能力扩展到更为广泛的各行各业，一波行业专用大模型在经过特殊的行业数据训练和参数调整即预训练-微调方法等复杂的技术熵变后应运而生，其中就包括好几个中国酒店住宿业的行业大模型：例如擅长住中体验的云迹大模型“HotelGPT”；专注酒店业知识管理和制度、标准的辉驿大模型“妙沿”；聚焦酒店数字营销的众荟大模型“通荟”;专攻酒店营运现场管理的首旅如家“数字店长”；而迈点AI大模型有点酒店业“万宝全书”的特点。
相信面对这些行业大模型绝大多数人有点瞠目结舌，一连串的疑惑浮现在深思的脸庞上：这些号称酒店行业大模型的东西和ChatGPT、文心一言、星火究竟是啥关系？科技公司如何训练出所谓的专业“大模型”？他们会在未来的酒店数字化过程中扮演什么角色，解决什么问题？如果真的有用，酒店集团和科技公司与它们最和谐的关系应该是如何？其未来的商务模式可能会是那种形式？
我关注和推进了其中几个大模型的研发过程，目睹了酒店数字专家在这个领域里辛勤耕耘的全过程，除了感动更有思考和殷切期待。 数字化转型网www.szhzxw.cn
一、酒店AI大模型和GPT等通用大模型是一脉相承的关系还是独立的创新？
ChatGPT及一众国内通用大模型出世后，人们尝到了AI的甜头：发现工作的思路、灵感被唤起不再是一种奢望；往日劳神费力的paperwork像制作PPT和Excel瞬间轻松起来，工作效率大大提高；就是非常烧脑的编码技术也从神坛上跌落下来，只要设想好一个模式，大模型就会迅速帮你输出一串令人心跳的Python代码或其他机器语言。不过，有心人很快发现，这些通用大模型在整体智能水平大幅度提高的情势下，用到专业领域或处理360行各等繁杂特殊的事务时智力立马下降，甚至于胡说八道。比如说让通用模型去处理酒店住宿业的许多专业问题它就显得力不从心，显然有高智商但无专业知识是这些通用模型的最大问题。
事实上ChatGPT大模型在推出时，OpenAI就预料到会有这样的情形，通用模型虽然涵盖了广泛的知识，但对于特定行业或领域的深入了解必定有所不足。这是因为每个行业特定的语境和特定的术语完全不同，知识体系和技术路线也天壤之别，只有通过专业训练才能帮助模型更好地理解和使用这些术语以解决行业问题。因此大模型广泛采用开源的方法，或将API开放给各个行业，相当于把GPT的脑袋借给需要的行业，通过对模型进行特定领域的训练，使其更精确地回答专业问题。这就是现在的通用模型衍生出专业（或行业）模型的底层逻辑。
例如这次联合发布的“酒店AI大模型”中，云迹科技的HotelGPT在百川2-13B、阿里千问-14B等基础上进行训练，辉驿科技的“妙沿”利用Langchain框架调用AI通用模型为“白泽”的内核引擎，众荟科技则在ChatGPT/ChatGLM等基础上进行研发和训练，首旅如家在GPT3.5/chatglm2基础上训练，迈点采用百度“文新千帆”作为训练底座。
这些行业的大模型无一例外地把自己的“脑子”建立在某一个通用大模型上，相当于借用了一个出色的“大脑”，但又无一例外地对“大脑”进行了各有侧重的强化训练，俗话说“恶补专业知识”，通过这样的环节“酒店AI大模型”有了自己的特殊本领，它成为了酒店住宿业的行家、专家。 数字化转型网www.szhzxw.cn
举一个例子，当系统从网络上爬取了一个客人对酒店的一段评论：“这家酒店的服务太热情了”。通常被归类好评：服务人员非常友好和周到，但在酒店业的语境里，它还有另一种可能：服务过于热情，客人可能感到有些骚扰，造成了不便，这就是抱怨了。酒店行业大模型要的是能够充分理解各种酒店住宿业特殊场景下的具体问题的处理能力。
因此“酒店AI大模型”并非独立门户、自开山头建立起来的一种AI系统，它是建立在多种通用大模型基础上，灌输以行业特种知识、技能训练使之更适合住宿业的一种专家系统。“酒店AI大模型”的推出，将改变以往的AI如“智能语音前台”或“某某智能音箱”回答问题时“专业性”较强（话题范围很窄），但一旦客人转移话题系统立刻变傻的现状。并且由于是建立在更强大的通用大模型基础上，其通用智能非常强大，这又促进了“酒店AI大模型”专业能力的极大提升。从这个意义上说，“酒店AI大模型”与市场上诸如ChatGPT一类的通用大模型有千丝万缕的联系，但从技术路线上来说，可能系出多门，未必一定扯上血缘关系。
那是否每个行业都要推出自己的专业大模型呢？
我以为未必。首先，对于许多一般的知识和查询，通用大模型如ChatGPT、盘古、星火可能就足够了，但对于需要精确或专业知识较“专”的行业（比如酒店、医疗、冶金等），创建专业大模型可能更有价值和市场。另一个原因是创建、训练和维持一个专业大模型需要长期的投入。
二、“酒店AI大模型”的出世对行业数字化意味着什么？
理论上说，“酒店AI大模型”的推出对行业数字化进程是一种“如虎添翼“的正面促进作用，实际效果怎样则要看大模型的真实功能和效用，即真本实事，以及行业对这些大模型的态度：拥抱还是冷漠。
在华邑酒店的咖啡房里，有着酒店行业人工智能领军人物之誉的林小俊—众荟科技CEO，给我分享了一段抖音直播视频：一位老练的酒店主播略带口音，正侃侃而谈推销着自己的套餐：放价不停，酒店特惠。一边介绍套餐的种种诱人的福利，一边鼓励观者线上提问，线上作答。小俊问我对主播的表现评价如何？回答：非常够格！小俊自豪地告诉我，那是众荟的大模型。
首旅如家是中国酒店业中第一个推出专业大模型的酒店集团，CIO王波起了关键推动作用。他把这个模型直接称为“数字店长”，是希望“它将每日的经营复盘、市场热点判断、酒店收益管理、宾客服务反馈、风险和应急处理等酒店60%的职能承担起来”。 数字化转型网www.szhzxw.cn
云迹负责大模型开发的曾祥永专家，是中科院技术研究所博士后，对我举例说他们的大模型让机器人地图更新、动态避障、路径规划、任务调度、人机交互等多方面技术指标和产品性能大幅度提高。云迹已有超过2万多家的酒店客户，为了处理机器人每天应用的各种问题，有一个数十人的售后客服团队，在大模型支撑下，售后服务效率得到了显著提升。在曾博士后方支持的云迹VP应甫臣，是Google中国第一代工程师，他对专业大模型的价值判断是：经历了早期的火热后，当下的关键在于真正提升各垂直领域的生产力。应总希望更多的大模型加入进来，一起加速酒店业数字化进程。
潘哲恺是辉驿科技专用大模型的主持者，这位毕业于美国纽约大学且立志要成为一名“数据科学家”的年轻人，对大模型风生水起创造的机会非常珍惜，对辉驿科技为他提供的舞台十分感激。他对辉驿大模型将给酒店住宿业带来的影响充满了自信，期待“辉驿的大模型将是未来酒店行业的”知乎”。
迈点AI的开发负责人张超阳，在大数据架构和AI模型设计上有丰富的经验。他特别自豪：得益于迈点14年来全域数据库资源，他们大模型花了整整6个月，持续从海量的行业新闻、研究报告中融合学习，站在酒店人维度分析，训练量超过100多万tokens,终于能够与人聊天结合上下文对话互动，高效快捷的帮助酒店人检索内容、创造解决方案。
这样对“酒店AI大模型”的点状评价有许多，且多层次、多侧面。
如果系统化地描述，那可以概括为以下几点：
⒈“酒店AI大模型”可以支持酒店日常营运，进行客房需求预测，物料预测、房价预测、收益预测，可以帮助管理酒店的房间存量和制定价格策略。帮助改善客户入住和退房流程，提高运营效率。
⒉“酒店AI大模型”可以用来分析市场趋势和客户反馈，以调整和改进酒店的营销策略。可以帮助生成个性化的电子邮件和营销材料，以吸引不同客户群体。
⒊“酒店AI大模型”可以进行良好的物业管理，可以改善各种设备、资产的运行情况，对各种运行的系统进行智能化的控制。 数字化转型网www.szhzxw.cn
⒋“酒店AI大模型”可以帮助酒店高效实现OA管理，包括各种文本翻译、文本起草、社交媒体的推文生成以保持活跃度。自动回复客户在社交媒体上的提问或评论，提供基本信息或处理客户问题。
⒌“酒店AI大模型”用于接受来自各个渠道的客户意见、需求，进行分析，按照分类予以反馈、并采取适当的行动来改进服务。这种接受和反馈的渠道和方式是融媒体全智能的。
⒍其他用途的支持。
总之“酒店AI大模型”对未来酒店数字化的推进是全方位、全场景、全过程、全智能的。具体的应用场景无法穷举，它是一个不断创新的过程。因此，“酒店AI大模型”是未来酒店住宿业数字化加速的“引擎”，但当下更像是与数字化一起脉动的探索者。
三、 从通用大模型到“酒店AI大模型”的浴火重生到底蕴含了哪些必须的流程？
不管基于何种通用大模型，要将它训练成为精通某一行业或某一领域知识或技能的专业大模型是一项高技术、复杂的、迭代性的过程。
我这里简单地将此过程的主要步骤描述一下：
首先要明确设计的目标。例如，如果目标是为酒店行业创建一个专业模型，那么应该考虑该行业中最常遇到的问题、需求以及客户的疑虑。还有酒店住宿业也分成许多领域，比如纯酒店服务、餐饮、娱乐、物业管理、运营、市场营销等等，在具体的领域里面还可以细分，如物业管理又可分为能源、设备和设施管理、安全和保安、清洁和环境卫生、园艺和景观、废物管理、停车管理、房务维修等等。
理论上可以做一个酒店住宿业的通用大模型，但这方方面面涉及知识还是太广、数据训练量巨大，投入也水涨船高。那么集中力量做一个领域“专业通”是比较靠谱的做法，现实的情况也是如此，这次发布的几个“酒店AI大模型”都有自己的特色。所谓特色，就是说在某个领域我特别行，言下在别的领域我还不怎么样。
如何确定大模型要达到的目标？常常和科技公司所涉及的领域和相关数据收集与处理的能力强势在哪个方面有关。 数字化转型网www.szhzxw.cn
重要的在于是否拥有数据收集与处理的优势。
对于酒店行业或其中细化领域（如住中行为、数字化营销、SOP和相关制度）的专业模型，需要大量与该酒店住宿业或领域相关的数据。这些数据包括文档、手册、行业报告、客户互动记录等。
举云迹的例子，2万多台机器人在全国各档次的酒店里整日跑来跑去执行各种任务，云迹积累了千万级的数据，这些数据有的是客人通过APP下达的购物指令，有的是客人通过电话或电视机下达的订餐购物需求，有的是前台发出的送物指令，有的是在电话中投诉，抱怨或表扬等等。这些数据统一经过云迹的HDOS处理，转换成最终的执行命令并进入云迹云端的大数据库，这就是云迹研发专业大模型的资源和底气。
再如众荟科技，十多年的舆情跟踪服务，他们的机器爬虫从各种融媒体上抓取了几十万家酒店基础信息和4亿条消费者的服务评价（俗称网评）。这些网评在经过众荟的精细化标签后裂变成细粒度的分析结果，成了人工智能系统辨析网评背后的情绪、语言真实的含义、行业特殊内涵的利器和资源。众荟利用这些数据资源，开发了以市场营销为主的大模型也就顺理成章了。
一般来说，有了数据，还应该对原始数据进行清洗、标注和分类，确保数据的质量和准确性，以便模型能够更好地理解和分析这些数据所含的信息。这包括对评论、反馈、房间描述、系统日志等文本进行清理和标记，假设您收集了一些客户对酒店的评论和反馈，这些评论可能是各种形式的文字，有些可能不够规范或混乱。在数据清洗和标记过程中，您会发现文本中的一些杂乱或不必要的部分，比如拼写错误、特殊字符、HTML标记、语言中个人习惯的“哼、哈”声或其他噪音。例如，将”!!!!!”替换为”!”，或者删除不必要的HTML标记，以确保文本更干净和一致，这就是数据清理。
而数据标记是将文本分成更小的有意义的部分的过程，以便大模型能够理解文本的结构和内容。在网评中，可以标记诸如评论主题、情感表达（积极、消极、中立）、提到的酒店设施等元素和评价维度等。这有助于大模型更好地理解评论的含义和信息。再通俗地说，数据清洗和标记就像是整理一本混乱的书籍，清洗就是去除书页上的涂鸦和杂乱的笔记，使它变得更容易阅读；标记就是在书中加入目录、章节标题和页码，以帮助读者更快地找到他们感兴趣的内容。通过清洗和标记，可以使大模型更好地理解和利用这些数据，提高其在酒店行业的应用效果。
经过上述准备后，还要进行最关键的一步，选择通用大模型：到底选ChatGPT,还是选国内的星火、盘古、百川等，这也是一个巨大的复杂工程，必须和你的信息系统技术环境最佳融合和符合你的人才支撑。因为这一步过于专业，此地略过。
然后将数据输入经过精心选择的通用大模型，通过大量的重复性的训练，就会积累对行业特有语境的敏感性和广泛的知识面，它已经具备了GPT之类的通用大模型没有的专业素养。此时还要对大模型进行“两环四节”的再调整性训练：微调和训练及验证和测试。 数字化转型网www.szhzxw.cn
前者是继续用特殊数据和案例对专业模型进行微调和再训练，促使大模型在特定的领域中表现更加优秀，而不是仅仅保持通用的知识。后者是将专业大模型放在实际场景中进行测试，以验证其性能和准确性。这包括利用行业专家提供的问题集，或使用真实的用户查询进行测试。这后面的“两环四节”进行得彻底与否，也会对专业大模型的最终使用效果产生直接的影响。
同时上面提到的“微调”是一种非常关键的训练途径，目的使专业大模型更适应特定工作的方法，换句话表达就是对专用模型进行一种个性化的培训，以便它更好地完成酒店业特定领域的工作。
举个例子，假设你训练”酒店AI大模型”，它在理解和回答一般性问题方面表现尚可，但你希望它能够在酒店业务中做更多的工作，比如提供酒店推荐、回答客户问题或者预测酒店房价。特定任务“微调”的过程就是让这个模型通过接受一些与酒店市场、预订、分销业务相关的数据，来学习如何更好地完成这些特定的任务。这些数据可以包括客户的酒店评价、酒店房价的历史数据、酒店出租率和房价的关系、酒店一般的收益管理原则、酒店设施的描述等等。通过在这些数据上进行训练，“酒店AI大模型”可以逐渐学会在酒店这些业务中有更好的表现，比如根据客户偏好推荐酒店，回答客户的问题，或者预测未来的价格趋势。
经过以上的繁复和大数据量的训练过程，专业大模型算是初具形态，但距离向行业推出的目标至少还有二个步骤要去完成，每一步都是伤筋动骨的历练：1、部署与实际应用；2、持续的学习和更新。
当然专业大模型的训练还有其它多种方法，例如众荟就采用了混合模型法、辉驿采用集成性和API能力较强的Langchain框架等等，这里我就不一一叙述了。
四、通用大模型和“酒店AI大模型”的商业模式的探索
通用大模型和专用大模型都是在人工智能底座上经过大量训练培养出来的一种公共和专用服务平台，其成长和维护必将耗用大量的人力、智力和财力，从本质上是一种高端服务。事实上也为使用者无论是机构、企业还是个人；系统还是直接的对人都有巨大的智力提升和劳效提升之功。因此，从长远看它的商务模式必然是有偿的。
简单说来目前大约有以下几种模式正在探索中：
1、订阅制/会员制：类似于许多在线服务和应用程序，用户为持续访问和获得最新、最优化的服务版本而支付费用。
2、按使用次数付费：每次请求、每分钟使用或每千字等等，都有不同的费用标准。对国内酒店行业大多数普通用户来说，使用“酒店AI大模型”是最方便的使用方式，为此付些费用情理上是合理的。
3、企业合作和定制化服务：企业可能需要特定的、定制化的大模型解决方案。这可能涉及在特定数据集上进行培训、优化性能或添加专有功能。这些服务比标准使用更为高级，因此价格也要高些。
4、开放API和开发者工具：开发者可能希望将这些大模型集成到自己的应用程序或系统中。通过提供有偿的API访问和开发者工具，扩大其大模型的应用范围并赚取收入。
5、免费的或付费的使用：大多数免费的是基础版本或有限制次数的版本，付费的常常是高级版本。如ChatGPT3.5是免费版，4.0是付费版。 数字化转型网www.szhzxw.cn
根据“2023酒店住宿业数字化发展峰会”联合发布会上得到信息，这次联袂推出的几家“酒店AI大模型”其普遍还处在试水阶段，商务模型并未成型：众荟采用License授权、辉驿显示为“内测”、云迹采用植入机器人增强性能、首旅如家为集团内部试用再视情推广、迈点为微信小程序免费订阅。
不管怎样，我预计，上面几种常用的商务模式对“酒店AI大模型”基本可选，问题在于我们酒店行业（酒店集团、独立酒店、科技公司、相关人员等）能否接受这种新型的数字化服务，并将它与自己的事业结合起来。只有当这种接受和结合呈现良性状态时，商务模式才可以被称作“合适”的。
五、“酒店AI大模型”与酒店集团和科技公司的和谐关系
前面我们阐述了“酒店AI大模型”对未来数字化进程具有强大的正向推进关系，那么是否所有的酒店集团、科技公司都有必要投入资金和人力去研发一个自有的“小宝贝”呢？
中国人有一个习惯，叫做“群羊效应”，形象地描述就是说好那就大家做，说不好全嫌弃。且不说大模型这东西真的不好做，就是资金和人力的投入也是很大的一本账。比较“和谐”的做法是“拿来主义”，为我所用。
酒店集团可以和这些“酒店AI大模型”建立直接的合作关系，将他们接入到自己的数字化系统中去，如接到CRS系统以增强客房的预订能力、接入到自己的物业系统中，强化自己的各种设备的管理能力、接入到自己的OR系统，优化自己的文档和知识管理的路径。零零总总，凡是现有的信息系统都可以和有特殊专长的AI大模型链接起来，仿佛为自己的系统添加了一个“AI智能引擎”，何乐不为？
对科技公司也是一个提升产品智能化的契机。目前酒店行业科技公司软硬件产品、系统已经相当丰富，但就其水准而言普遍停留在信息化时代，也就是以自动化为特征的“智能酒店1.0时代”，各种产品“自动”可以，智能不足。并且令大多数科技公司和CTO烦恼的是，产品开发似乎已经走进了山穷水尽疑无路的死胡同，不知突围的方向在哪里?今年七月，我在青岛为首旅如家的数字化专家讲授《酒店科技企业突围的“五个维度”》时就指出，重要的维度之一就是“让现有的产品更智慧”，而更智慧的捷径之一就是引入“酒店AI大模型”。 数字化转型网www.szhzxw.cn
也许对以上观点大多数同仁是赞成的，但也一定会有人这样想：既然“酒店AI大模型”对数字化建设这么有用，为何我自己不也搞一个？这样的想法当然不能算错，却不一定明智。一方面，有进入“群羊效应”的嫌疑，一哄而上的结局必然是大面积淘汰。另一方面，“酒店AI大模型”虽建立在通用大模型上面，然而把“通用大模型”训练成“专用”大模型的过程绝不是喝着咖啡看看我文章解释那么轻松，几乎每个“酒店AI大模型”的背后都是博士级数字化专家呕心沥血的过程，我将之称为“凤凰涅槃-浴火重生”可谓言之凿凿，充满敬意。
也有的同仁会说，引入“AI引擎提高自己系统的智慧性”这个说法我接受，但为什么不直接引入ChatGPT或国内的星火、盘古？选择性更多，使用体验也不错。为什么要引进或鼓励使用““酒店AI大模型”？这种说法我在与诸多CIO和CEO交谈中多次听到，可以说有一定的代表性，但也显示出这些观点的持有者对大模型的认识还是肤浅的，对行业内的“合作与共识”建立的观念是淡薄的，我明确表示不敢苟同。
说这种观点是肤浅的，是基于现在推出的几款“酒店AI大模型”的数字底座都是国外或国内的通用大模型，瞧不起这些行业大模型的基本能力就是瞧不起这些能力的提供者，诚如我在本文第三节中提到，这些“酒店AI大模型”就是在这些通用模型基础上通过大量行业数据训练而得到的，可以怀疑这里的专业训练是否满足你的特别要求，也就是说可以通过使用它，去质疑它的专业AI能力是否足够强，并且用自己的数据或场景帮助它不断迭代、进步是比较好的做法。而一方面否定“酒店AI大模型”的底座智能性，一方面去引进被自己怀疑能力的通用大模型底座，这可以成为一种做法，却是一种违背逻辑的做法。
说它是一种对行业内的“合作与共识”的淡薄观点，酒店住宿业本来就是一个传统服务行业，在数字化方面的人才和资源特别珍贵、这几年疫情又使行业受到重创，其数字化进程落后于其他行业，也落后于国家数字化建设规划的阶段目标。我们最好的办法是集中优势资源，各干各的强项，同时用别人的强项或成果，形成一种合作环境，并将这种合作成为共识。
酒店集团和独立酒店、科技公司和酒店各路专家，与“酒店AI大模型”加紧联系，尽可能多地在各种场景中使用，其实也是一种最佳的“再训练”。大模型有不断学习的功能，它在提供服务的同时，也在吸取使用者的反馈。一方面使用者的场景是丰富的，是大模型研发者无法穷尽的，因此在大模型的数据库里必然存在一定的盲区；另一方面，你的使用及对结果的评价对大模型逻辑推理和算法都有不断完善的的推力，可以最大限度克服大模型极有可能存在的“偏见”。因此从这个意义上说“使用”大模型本身就是一种对“酒店AI大模型”的支持和贡献。
山东舜和酒店集团的VP任丛丛对此这样评价，她感觉这些大模型一旦用到酒店的系统中，在美丽的环境、舒适的客房和善解人意的大模型支持下的各种智能系统，能够准确理解客人的需求，并提供个性化、高效的服务体验，这绝对是酒店业引领数字化时代的闪亮高光点。她乐观地预测这会“让我们的酒店更具竞争力和魅力！” 数字化转型网www.szhzxw.cn
我相信，像任丛丛这样对“酒店AI大模型”情有独钟，热情拥抱的酒店住宿业高管不在少数。他们会是一群园丁，用自己的智慧去浇灌、培养这些刚刚出土的幼苗；他们所在的酒店、公司、机构将会变成肥沃的土壤，让“酒店AI大模型”在使用中茁壮成长。
诚然，我对这些“酒店AI大模型”的最终效果并不敢轻率评论或虚言妄赞，因为我和大家一样还没有足够的使用经历，但我愿意多多地使用他们、支持他们、完善他们，其中虽有淘汰者也必有成功者。
这就是中国饭店协会酒店数字化专委会和我本人希望推进建立的“和谐”关系。
英文翻译：
“Suddenly like the spring breeze of the night, thousands of trees pear blossom” this wonderful poem is very appropriate to describe the emergence of the current hotel industry model. On October 27, at the joint press conference of the “2023 Hotel Accommodation and Catering Industry Digital Development Summit” held, Yuntrace, Huyi, Zhonghui, Beijing Travel Home, and Mai Dian jointly released their respective “hotel AI big model”, which was really a little dizzying for a while.
This is the closest China’s hotel industry has ever come to embracing new digital technology.
At the end of last year, OpenAI’s ChatGPT was born, setting off a wave of artificial intelligence frenzy. And people were shocked by the extraordinary ability of the AGI large model. In early March of this year, the Hangzhou Secret Service invited me to give a lecture to the executives of the West Lake State Guesthouse, specifically requesting the addition of ChatGPT content. I used ChatGPT3.0 to automatically write a set of PPT introducing the ins and outs of ChatGPT. And gave an example explaining the advantages and disadvantages of GPT. To be honest, ChatGPT’s overall strength, despite its serious nonsense, shocked me, and I realized that once the hotel industry uses this so-called “big model”, it will undoubtedly enhance the digital process. 数字化转型网www.szhzxw.cn
In the short period of six months that followed, the domestic general large model developed rapidly, like mushrooming. Baidu’s “Wenxin Word”, Ali’s “Tongyi thousand questions”, Huawei’s “Pangu”. IFlytek’s “Spark”, China Mobile’s “nine days of engines”, China Unicom’s “Honghu” . And so on dozens of large models have been born. Although the performance, compared with ChatGPT4.0 some distance. But in the overall function is also very useful, with time. And after a lot of Chinese material training, in quality to catch up with foreign products quite hopeful.
The emergence of ChatGPT and a large number of domestic general large models has extended its capabilities to a wider range of industries with its open source characteristics and spillover effects.
A wave of industry-specific large models came into being after complex technical entropy changes such as special industry data training and parameter adjustment, that is, pre-training-fine-tuning methods. Among them, there are several large models of China’s hotel accommodation industry: for example, the cloud track large model “HotelGPT” which is good at staying in the hotel experience; Focus on the hotel industry knowledge management and systems, standards of the big model “Miaoyan”; Focus on the hotel digital marketing model “Hui”; Specializing in hotel operation site management of the first travel home “digital store manager”; The Mai Point AI large model has the characteristics of the hotel industry “Wanbao encyclopedia”.
I believe that the vast majority of people in the face of these industry models are a little dumbstruck. And a series of doubts emerge on the thoughtful face: What is the relationship between these so-called hotel industry models and ChatGPT, Wenxin Word, and spark? How do tech companies train so-called “big models” of expertise? What role will they play and what problems will they solve in the digital future of hotels? If so, what would be the most harmonious relationship between hotel groups and tech companies with them? What is the business model likely to look like in the future?
I paid attention to and promoted the development process of several large models, witnessed the whole process of hard work by hotel digital experts in this field. And in addition to being moved, I was more thoughtful and eager.
First, is the hotel AI grand model and GPT general grand model a consistent relationship or independent innovation?
After the birth of ChatGPT and a number of domestic general models, people have tasted the sweetness of AI. it is no longer a luxury to find work ideas and inspiration to be aroused .Paperwork, such as making PPT and Excel, which used to be laborious, was instantly relaxed. And work efficiency was greatly improved. Even very brain-burning coding techniques have fallen from the altar, as long as you imagine a good pattern, the big model will quickly help you output a string of heart-pounding Python code or other machine language. However, people soon found that these general large models in the overall level of intelligence greatly improved. The use of professional fields or 360 lines of various complex special affairs intelligence immediately decreased, and even nonsense. For example, the general-purpose model is inadequate to deal with many specialized problems in the hotel accommodation industry, and it is clear that having high intelligence but no specialized knowledge is the biggest problem of these general-purpose models.
In fact, when the ChatGPT grand Model was introduced, OpenAI anticipated that the general model, while covering a wide range of knowledge, would necessarily fall short of the in-depth understanding of a particular industry or domain.
This is because the specific context and specific terminology of each industry are completely different. And the body of knowledge and technical routes are also very different. And only professional training can help models better understand and use these terms to solve industry problems.
Therefore, the wide adoption of open source methods for large models, or the opening of apis to various industries, is equivalent to lending the brains of GPT to the industry in need, by training the model in a specific domain, so that it can answer professional questions more accurately. This is the underlying logic behind the derivation of today’s general-purpose models into specialized (or industry) models.
For example, in the “Hotel AI big model” jointly released this time, Yuntrace Technology’s HotelGPT is trained on the basis of Baichuan 2-13B and Ali Qianwen-14B . And Huyi Technology’s “Miaoyan” uses the Langchain framework to call the AI general model “Bai Ze” core engine. Zhonghui Technology conducts research and development and training on the basis of ChatGPT/ChatGLM, while the First Travel Home trains on the basis of GPT3.5/chatglm2, and the Maipoint uses Baidu “Wenxin Qianfan” as the training base.
The large models of these industries without exception build their own “brain” on a general large model, equivalent to borrowing an excellent “brain”.
But without exception, the “brain” has been focused on intensive training, as the saying goes. “bad professional knowledge”, through such links, “hotel AI large model” has its own special skills. It became a connoisseur and expert in the hotel accommodation industry.
For example, when the system crawls a review of a hotel from the Internet by a guest: “The service of this hotel is too warm”. Usually classified as positive: the service staff is very friendly and attentive. But in the context of the hotel industry. It has another possibility: the service is too warm, the guest may feel a bit harassed, causing inconvenience, which is a complaint. The large model of the hotel industry requires the ability to fully understand the specific problems in the special scenarios of the hotel accommodation industry. 数字化转型网www.szhzxw.cn
Therefore, the “hotel AI Grand model” is not an independent portal.
And an AI system established from the beginning of the mountain. It is an expert system that is based on a variety of general large models and instills industry special knowledge and skill training to make it more suitable for the accommodation industry. The launch of the “hotel AI big model” will change the past AI such as “intelligent voice reception” or “certain smart speaker” when answering questions is “professional” (the topic range is very narrow), but once the guest changes the topic system immediately becomes silly.
And because it is built on the basis of a more powerful general large model. Its general intelligence is very powerful, which has promoted the professional ability of “hotel AI large model” greatly. In this sense, the “hotel AI large model” is inextricably linked to the general large model such as ChatGPT on the market. But from the technical route. It may be a lot of doors, not necessarily related to the blood relationship.
Does every industry have to come up with its own big model?
I don’t think so. First of all, for many general knowledge and queries, general-purpose large models such as ChatGPT, Pangu, Spark may be sufficient. But for industries that require precision or more “specialized” knowledge (such as hospitality, healthcare, metallurgy, etc.), creating specialized large models may be more valuable and marketable. Another reason is the long-term commitment required to create, train, and maintain a large professional model.
Second, what does the birth of the “hotel AI grand model” mean for the digitalization of the industry?
In theory, the launch of the “hotel AI big model” is a positive promotion of the industry digitalization process. And the actual effect depends on the real function and utility of the big model, that is, the real facts, and the industry’s attitude towards these big models: embrace or indifference.
In the coffee room of Huayi Hotel, Lin Xiaojun, the CEO of Zhonghui Technology, who has a reputation as a leader in artificial intelligence in the hotel industry, shared a live video of Tiktok with me: a seasoned hotel anchor with a slight accent, is talking and selling his package: the price is not stop, the hotel special offer. While introducing the various attractive benefits of the package, viewers are encouraged to ask questions and answer online. Xiaojun asked me how to evaluate the performance of the anchor. Answer: Very qualified! Xiao Jun proudly told me that it was a large model of Hui.
Home Inns was the first hotel group in the Chinese hotel industry to launch a professional large-scale model, and CIO Wang Bo played a key driving role. He calls this model directly “digital store manager” and hopes that “it will assume 60% of the functions of the hotel such as daily business review, market hot spot judgment, hotel revenue management, guest service feedback, risk and emergency handling.”
Zeng Xiangyong, an expert in charge of large model development at Yuntrace. And a postdoctoral fellow at the Institute of Technology. Chinese Academy of Sciences, told me that their large model has greatly improved the technical indicators and product performance of robot map update, dynamic obstacle avoidance, path planning, task scheduling, human-computer interaction and other aspects. Yuntrace has more than 20,000 hotel customers, in order to deal with the various problems applied by robots every day.
There is a after-sales customer service team of dozens of people, with the support of the large model.
The after-sales service efficiency has been significantly improved. In the rear support of Dr. Zeng Yun trace VP Ying Fuchen, is the first generation of Google China engineers, his value judgment of the professional large model is: after the early hot, the key is to really improve the productivity of each vertical field. We hope that more large models will join us to accelerate the digitization process of the hotel industry.
Pan Zhekai is the host of the special large model of Fuyi Technology. The young man who graduated from New York University. And aspires to become a “data scientist” cherishes the opportunity of creating a large model very much. And appreciates the stage provided by Fuyi Technology for him. He is full of confidence in the impact of the grand model of Fuyi on the hotel accommodation industry. And expects that “the grand model of Fuyi will be the” knowledge of the hotel industry in the future “.
Zhang Chaoyang, the development director of Maidian AI, has rich experience in big data architecture and AI model design. He was particularly proud of: Thanks to the 14 years of global database resources, their big model spent a full 6 months, continued to integrate learning from massive industry news. And research reports, stood in the hotel human dimension analysis, training amount of more than 1 million tokens, and finally able to chat with people combined with contextual dialogue interaction, efficient and efficient help hotel people to retrieve content and create solutions.
In this way, there are many point evaluations of the “hotel AI big model”, and they are multi-level and multi-sided.
If described systematically, it can be summarized as the following points:
“Hotel AI large model” can support the daily operation of the hotel, room demand forecast, material forecast, room price forecast, income forecast, can help manage the hotel room stock and formulate price strategy. Help improve customer check-in and check-out processes to improve operational efficiency.
The “Hotel AI Grand Model” can be used to analyze market trends. And customer feedback in order to adjust and improve the hotel’s marketing strategy. Can help generate personalized emails and marketing materials to appeal to different customer groups.
The “hotel AI large model” can carry out good property management, improve the operation of various equipment and assets. And carry out intelligent control of various operating systems.
The “Hotel AI Grand Model” can help hotels efficiently achieve OA management, including various text translation, text drafting. And Twitter generation on social media to stay active. Automatically respond to customer questions or comments on social media, provide basic information or deal with customer questions.
The “Hotel AI big model” is used to accept customer opinions. And needs from various channels, analyze them, give feedback according to classification. And take appropriate actions to improve the service. This kind of reception and feedback channels and ways are fully intelligent. 数字化转型网www.szhzxw.cn
In short, the “hotel AI grand model” is an all-round, full-scene, whole-process, and fully intelligent promotion of future hotel digitalization. Specific application scenarios can not be exhaustive, it is a process of continuous innovation. Therefore, the “hotel AI big model” is the “engine” of the digitalization acceleration of the hotel accommodation industry in the future. But the current is more like an explorer pulsating with digitalization.
Third, what are the necessary processes contained in the rebirth of the general large model to the “hotel AI large model”?
No matter what kind of general grand model is based on, it is a highly technical, complex. And iterative process to train it into a professional grand model that is proficient in the knowledge or skills of a certain industry or a certain field.
Let me briefly describe the main steps of this process:
First of all, define the goal of the design. For example, if the goal is to create a professional model for the hospitality industry. You should consider the most frequently encountered problems, needs. And customer concerns in the industry. The hotel accommodation industry is also divided into many areas, such as pure hotel services, catering, entertainment, property management, operations, marketing, etc.. In specific areas can also be subdivided, such as property management can be divided into energy, equipment. And facilities management, safety and security, cleaning and environmental health, gardening. And landscape, waste management, parking management, room maintenance and so on.
In theory, a general model of the hotel accommodation industry can be made. But the knowledge involved in all aspects is still too wide, the amount of data training is huge. And the investment is also rising. So focus on doing a field of “professional pass” is a more reliable approach, the reality is also the case. The release of several “hotel AI large model” have their own characteristics. The so-called characteristic means that I am very good in one area, but I am not good in other areas.
How to determine the goals to be achieved by the large model? It is often related to the fields involved in technology companies. And the strong ability of relevant data collection and processing.
What matters is whether you have the advantage of data collection and processing.
Specialized models for the hospitality industry or specific areas within it, such as in-residence behavior, digital marketing, Sops. And related systems, require a large amount of data related to the hospitality industry or sector. This data includes documents, manuals, industry reports, customer interaction records, and more.
Yuntrace, for example, has more than 20,000 robots running all day long in hotels of all grades across the country to perform various tasks. Yuntrace has accumulated tens of millions of data, such as shopping instructions issued by guests through apps, food ordering. And shopping needs issued by guests through phone or TV, delivery instructions issued by the front desk. And complaints made over the phone. Complain or praise, etc. These data are uniformly processed by HDOS of cloud trails, converted into final execution commands. And entered into the large database of cloud trails Cloud, which is the resource. And base of professional large model of cloud trails R & D. 数字化转型网www.szhzxw.cn
Another example is Zhonghui Technology, more than ten years of public opinion tracking services.
Their machine crawlers from various financial media to grab hundreds of thousands of hotel basic information and 400 million consumer service evaluation (commonly known as network review). These online reviews are divided into fine-grained analysis results after the refined labels of Crowdfeed, which become a tool and resource for artificial intelligence systems to distinguish the emotions behind online reviews, the real meaning of language, and the special connotation of the industry. With these data resources, it is only logical that Zhonghui has developed a large model based on marketing.
In general, with the data, the original data should also be cleaned, labeled, and classified to ensure the quality. And accuracy of the data so that the model can better understand and analyze the information contained in the data. This includes cleaning and tagging the text of reviews, feedback, room descriptions, system logs, etc., assuming you have collected a number of customer reviews and feedback about the hotel, which may be in various forms of text, some may not be normative or confusing. During the data cleansing and markup process, you will find some cluttered or unnecessary parts of the text, such as typos, special characters, HTML tags, the customary “hum, ha,” or other noise in the language. For example, to “!!!!!” Replace with “!” Or remove unnecessary HTML markup to ensure that the text is cleaner and more consistent, which is called data cleansing. 数字化转型网www.szhzxw.cn
Data markup, on the other hand, is the process of dividing text into smaller, meaningful parts so that larger models can understand the structure.
And content of the text. In online reviews, you can mark elements such as the subject of the review. The expression of emotion (positive, negative, neutral). The hotel facilities mentioned, and the dimensions of the review. This helps the larger model better understand the meaning and message of the comments. In layman’s terms, data cleaning and tagging is like organizing a messy book.
Cleaning is removing graffiti and cluttered notes from the pages to make it easier to read. Tagging is the addition of table of contents, chapter titles.
And page numbers to a book to help readers find what they are interested in more quickly. By cleaning and labeling, large models can better understand. And use these data to improve their application in the hotel industry. 数字化转型网www.szhzxw.cn
After the above preparation, but also to carry out the most critical step, choose the general large model: in the end to choose ChatGPT, or choose the domestic spark, Pangu, Baichuan, etc.. This is also a huge complex project. Must be the best integration with your information system technology environment and meet your talent support. Because this step is too technical, I’ll skip it here.
The data is then fed into a carefully selected general-scale model. And through a lot of repetitive training, it accumulates industry-specific context-sensitive. And broad knowledge, which already has the expertise that a general-scale model like GPT does not. At this time, the large model also needs to carry out “two rings . And four sections” readjust training: fine tuning and training and verification and testing.
The former is to continue to fine-tune and retrain specialized models with specific data. And cases to make larger models perform better in specific domains, rather than just maintaining generic knowledge. The latter is to test professional large models in real scenarios to verify their performance and accuracy. This includes using problem sets provided by industry experts, or testing using real user queries. Whether the latter “two rings and four sections” are carried out thoroughly or not will also have a direct impact on the final use effect of professional large models.
At the same time, the “fine-tuning” mentioned above is a very critical training approach to make the professional model more suitable for a specific job, in other words.
A personalized training of the special model so that it can better perform the work of the specific area of the hospitality industry.
For example, let’s say you train the Hotel AI Grand Model, which is good at understanding and answering general questions. But you want it to be able to do more in the hotel business, such as making hotel recommendations, answering customer questions, or predicting hotel rates. The process of “fine-tuning” specific tasks is for the model to learn how to perform those specific tasks better by receiving some data related to the hotel marketing, booking, and distribution operations.
This data can include customer reviews of hotels, historical data on hotel rates, the relationship between hotel occupancy rates. And rates, general revenue management principles for hotels, descriptions of hotel facilities, and so on. By training on this data, the “Hotel AI Grand Model” can gradually learn to perform better in these parts of the hotel business, such as recommending hotels based on customer preferences, answering customer questions, or predicting future price trends.
After the above complex and large data training process, the professional large model is beginning to take shape. But there are at least two steps to complete the goal of launching to the industry, each step is the experience of the bones: 1, deployment and practical application; 2. Continuous learning and updating.
Of course, there are many other methods for training professional large models, such as the mixed model method used by Zhonghui, the Langchain framework with strong integration and API ability used by Huiyi, and so on, I will not describe it here. 数字化转型网www.szhzxw.cn
Fourth, explore the business model of general grand model and “hotel AI grand model”
General large model and special large model are a kind of public and special service platform after a lot of training on the base of artificial intelligence, and its growth and maintenance will consume a lot of manpower, intelligence and financial resources, which is essentially a high-end service. In fact, it is also for users whether they are institutions, enterprises or individuals; The system or directly to people have a huge intellectual improvement and labor efficiency improvement. Therefore, its business model must be paid in the long run.
In short, there are currently about the following models being explored:
1, Subscription/membership: Similar to many online services and applications, users pay for continuous access to and access to the latest, optimized version of the service.
2, pay according to the number of use: each request, per minute use or per thousand words, etc., there are different fees. For most ordinary users of the domestic hotel industry, the use of “hotel AI large model” is the most convenient way to use. And it is reasonable to pay some fees for this.
3, Enterprise collaboration and customized services: Enterprises may need specific, customized large-scale model solutions. This may involve training on a particular dataset, optimizing performance, or adding proprietary features. These services are more advanced than standard use and therefore more expensive.
4.Open apis and developer tools: Developers may want to integrate these large models into their own applications or systems. Expand the reach of its larger model and earn revenue by providing paid API access and developer tools. 数字化转型网www.szhzxw.cn
5, free or paid use: most of the free version is the basic version or a limited number of versions. And the paid version is often the premium version. For example, ChatGPT3.5 is a free version, 4.0 is a paid version.
According to the information obtained at the joint press conference of the “2023 Hotel Accommodation Industry Digital Development Summit”.
The several “hotel AI large models” jointly launched this time are generally still in the water testing stage, and the business model has not been formed: Zhonghui is authorized by License, Huiyi is displayed as “private test”, Yuntrace uses implanted robots to enhance performance, BTT Home is used for internal trial and promotion according to the situation, and Mai Dian is a wechat mini program for free subscription.
In any case, I expect that the above common business models are basically optional for the “hotel AI grand model”, the question is whether our hotel industry (hotel groups, independent hotels, technology companies, stakeholders, etc.) can embrace this new digital service and integrate it with their own business. The business model can only be called “appropriate” if this acceptance and integration is in a benign state.
Fifth, the harmonious relationship between “hotel AI big model” and hotel groups and technology companies
Previously, we explained that the “hotel AI big model” has a strong positive relationship to the future digital process, so whether all hotel groups and technology companies are necessary to invest money . And manpower to develop their own “little baby”? 数字化转型网www.szhzxw.cn
The Chinese have a habit, called the “flock effect”, which is described in an image as saying that if it is good, everyone will do it, and if it is not good, it will be rejected. Not to mention that the big model is really difficult to do, that is, the investment of capital and manpower is also a big account. The more “harmonious” approach is to “bring the doctrine” for my use.
Hotel groups can establish a direct cooperative relationship with these “hotel AI grand models” and connect them to their own digital systems, such as connecting to the CRS system to enhance the booking ability of rooms, connecting to their own property systems, strengthening their own management capabilities of various devices, connecting to their own OR systems, and optimizing their own documentation and knowledge management paths. All the existing information systems can be linked with a large AI model with special expertise, as if adding an “AI intelligent engine” to their own system, why not?
It is also an opportunity for technology companies to improve product intelligence. At present, the software and hardware products and systems of technology companies in the hotel industry have been quite rich, but in terms of their level, they generally stay in the information age, that is, the “intelligent hotel 1.0 era” characterized by automation. And various products can be “automatic” and lack of intelligence.
And to the annoyance of most tech companies and Ctos.
Product development seems to have reached a dead-end, where is the direction to break out? In July this year, when I taught the “Five Dimensions” of Hotel technology Enterprise Breakthrough to the digital experts of Home Inn in Qingdao, I pointed out that one of the important dimensions is to “make existing products smarter”. And one of the shortcuts to be smarter is to introduce “hotel AI big model”.
Perhaps most colleagues agree with the above views, but there will also be some people who think: since the “hotel AI grand model” is so useful for digital construction, why don’t I make one myself? Such thinking is not wrong, of course, but not necessarily wise. On the one hand, there is the suspicion of entering the “flock effect”. And the outcome of the rush is bound to be a large area of elimination. On the other hand, although the “hotel AI grand model” is built on the general grand model, the process of training the “general grand model” into a “special” grand model is not so easy to drink coffee and read my article to explain, almost every “hotel AI grand model” is behind the doctoral level digital experts painstaking process. I call it “the Phoenix rises from the ashes,” and I do it with all due respect.
Some colleagues will say that I accept the idea of introducing “AI engines to improve the intelligence of their own systems”.
But why not directly introduce ChatGPT or domestic spark and Pangu? More options and a good experience. Why introduce or encourage the use of the “Hotel AI Grand Model”? This statement, which I have heard many times in conversations with many CIOs and ceos, can be said to be representative, but it also shows that the holders of these views have a shallow understanding of the big model, and the concept of “cooperation and consensus” in the industry is weak, and I clearly disagree.
To say that this view is superficial, is based on the launch of several “hotel AI big model” digital base are foreign or domestic general large models, looking down on the basic capabilities of these industry big models is looking down on the providers of these capabilities, as I mentioned in the third section of this article.
These “hotel AI big model” is based on these general models through a large number of industry data training.
You can doubt whether the professional training here meets your special requirements, that is to say. You can use it to question whether its professional AI ability is strong enough. And with their own data or scenes to help it iterate, progress is a better practice. On the one hand, denying the intelligence of the base of the “hotel AI large model”, and on the other hand, introducing the universal large model base that is suspected by oneself, which can become a practice, is a kind of illogical practice.
It is a weak view of the “cooperation and consensus” in the industry, the hotel accommodation industry is a traditional service industry, in the digital aspect of the talent.
And resources are particularly precious, in recent years, the epidemic has made the industry hard hit, its digital process lags behind other industries, but also behind the national digital construction planning stage goals. Our best way is to concentrate superior resources, do their own strengths, while using others’ strengths or achievements, to form a cooperative environment. And this cooperation will become a consensus.
Hotel groups and independent hotels, technology companies and hotel experts, step up contact with the “hotel AI big model” and use it in as many scenarios as possible, which is actually the best “retraining”. The large model has a continuous learning function, it provides services, but also draws feedback from users. On the one hand, the user’s scenes are rich, which can not be exhausted by the developers of large models, so there must be some blind spots in the database of large models. On the other hand, your use and evaluation of the results have a push for continuous improvement of large model logic reasoning and algorithms. And can maximize the “bias” that is likely to exist in large models. Therefore, in this sense, “using” the large model is itself a support and contribution to the “hotel AI large model”.
Ren Cong Cong, VP of Shandong Shunhe Hotel Group, comments on this.
She feels that once these large models are used in the hotel system, various intelligent systems supported by beautiful environment, comfortable rooms and understanding large models can accurately understand the needs of guests. And provide personalized and efficient service experience, which is definitely a shining highlight of the hotel industry leading the digital era. She optimistically predicted that this would “make our hotels more competitive and attractive!”
I believe that there are not a few executives in the hotel accommodation industry like Ren Cong Cong who have a soft spot for the “hotel AI big model” and warmly embrace it. They will be a group of gardeners, with their own wisdom to water, cultivate these newly unearthed seedlings; Their hotels, companies, and institutions will become fertile soil for the “hotel AI grand Model” to thrive in use.
Admittedly, I do not dare to comment on the final effect of these “hotel AI big models”, because I do not have enough use experience like everyone else, but I am willing to use them more, support them, and improve them, although there will be winners. 数字化转型网www.szhzxw.cn
This is the “harmonious” relationship that the Hotel Digital Committee of the China Hotel Association and I hope to promote.
本文由数字化转型网（www.szhzxw.cn）转载而成，来源于新工业网；编辑/翻译：数字化转型网宁檬树。
免责声明: 本网站(http://www.szhzxw.cn/)内容主要来自原创、合作媒体供稿和第三方投稿，凡在本网站出现的信息，均仅供参考。本网站将尽力确保所提供信息的准确性及可靠性，但不保证有关资料的准确性及可靠性，读者在使用前请进一步核实，并对任何自主决定的行为负责。本网站对有关资料所引致的错误、不确或遗漏，概不负任何法律责任。
本网站刊载的所有内容(包括但不仅限文字、图片、LOGO、音频、视频、软件、程序等) 版权归原作者所有。任何单位或个人认为本网站中的内容可能涉嫌侵犯其知识产权或存在不实内容时，请及时通知本站，予以删除。
