Internet companies have a large number of lines of data, and the amount of data is growing fast, in addition to using big data to improve their business outside the Internet companies have begun to realize operational data, the use of Big Data to discover new business value.
Alibaba, for example, it not only continues to strengthen personal recommendation, "thousands of thousand faces" This consumer-oriented large data applications, and also trying to take advantage of big data intelligent customer service, this scenario will gradually from extend internal applications to external call center in many companies.
In the business-oriented big data applications to "business consultant" for example, more than 6 million businesses in the use of "business consultant" to enhance their electrical store operators face level. In addition to facing their own ecological outside, Alibaba operational data are accelerating, "Credit Sesame" This conduct personal credit rating is gained rapid development based on personal data collection, application scenarios extend from the inside to the Alibaba An increasing number of external scenarios, such as car rental, hotels, visas and so on.
Since all behavior of customers will leave marks on the Internet platform, so Internet companies can easily get a lot of customer behavior information. Information generated by the Internet business platform generally have authenticity and certainty through the use of Big Data technologies to analyze these data, can help companies develop targeted services strategy, in order to gain greater efficiency. In recent years the practice has proved that reasonable use of Big Data technologies can improve the operating efficiency of more than 60% of e-commerce.
Big Data in the past few years has changed the face of e-commerce, specifically, e-commerce industry Big Data applications in the following areas: precision marketing, personalized service, merchandise personalized recommendations.
1. Precision Marketing
Internet companies with large data collection techniques related to various types of customer data, and the establishment of "user-portrait" by big data analysis to describe abstract picture of a user's information, so the user can personalize the recommendation, precision marketing and advertising and so on.
When users visit the website of the moment, the system can predict the user today at why, and then from the library to the appropriate commodity commodity to find out, and recommended to him. Figure 1 shows the basic information which users and user properties will include portraits.
1 user portrait
Marketing at the core of large data support that make the company's business at the right time, through a suitable carrier, in an appropriate manner, pushed to most business users need this.
First, big data marketing has a strong timeliness. In the Internet era, users can easily change consumer behavior in a short time, big data marketing marketing strategy can be implemented in a timely manner when a user needs the most productive.
Second, it can implement personalized, differentiated marketing. Big Data marketing based on user interests, needs at some point in time, so that one marketing segment of users, let's do targeted marketing business, and can effect real-time feedback, timely adjustment Marketing strategy.
Finally, the big data marketing to target user's information can be associated analysis. Big data can be associated multi-dimensional analysis of all kinds of information the user found interesting associations and the relevant links between data items from large data sets.
For example, by discovering the link between the user's shopping basket of different commodities, analyze spending habits of other users. By knowing which items are purchased at the same time users frequently to help marketers a commodity from the user's consumption habits, found that users of another commodity consumption patterns, in order to develop a marketing strategy related products for this user. Figure 2 shows the site will recommend different products for different customers based on user portrait.
Figure 2 Precision Marketing
For example, an e-commerce platform for customers to browse through the network records and purchase records grasp customer consumption patterns, to analyze consumer-related characteristics and classification of customers. Such as income, family characteristics, buying habits, and ultimately master customer profile, and based on these features may be of interest to determine their products and services.
Consumers enter the site from the beginning, the site in the list page, single product page, shopping cart page and other four pages, deployed five columns of different algorithms recommended for applications recommended product of interest, thereby enhancing product visibility, promote cross and up-selling. After making a comprehensive optimization from multiple angles, the mall made up order conversion rate grew by 66.7%, made up of goods grew by 18 percent conversion rate, total sales increased by 46%.
Wal-Mart supermarkets in the United States, when the cashier finished scanning the customer purchase of goods, will show some additional information on the POS machine and the salesman will remind the customer based on the information which products can also be purchased. Wal-Mart achieved in large data systems to support "marketing consultant" system to build predictive models, for example, if a customer's shopping cart in a lot of beer, wine and salad, with 80 percent of likely need to buy wine with side dishes, condiments .
2. Personalized service
E-commerce has provided personalized service inherent advantages can be obtained online users through real-time technical support and timely provision of customized services for them.
许多电商都已经尝试了依靠数据分析，在首页为用户提供全面的个性化的商品推荐。海尔和天猫提供了让用户在网上定制电视的功能，顾客可以在电视机生产以前选择尺寸、边框、清晰度、能耗、颜色、接口等属性，再由厂商组织生产并送货到顾客家中。这样的个性化服务受到了广泛欢迎。
类似的定制服务还岀现在空调、服装等行业。这些行业通过满足个性化需求使顾客得到更满意的产品和服务，进而缩短设计、生产、运输、销售等周期，提升商业运转效率。
企业要为用户提供理想的个性化服务，首先必须通过数据充分了解用户的个性，其次是合理地掌控和设计服务的个性。了解用户个性是为用户提供他们想要的产品和服务的基础。企业需要在庞大的数据库中，找出最具有含金量的数据，然后，通过数据挖掘方法对用户进行聚类，再依据用户类型的特征设计针对性的服务。
个性化分散的单位可大可小，大到一个有同样需求的客户群体，小到每一个用户都是一个个性化需求单位。企业必须掌握好个性化服务的粒度，过于分散的个性化服务，会增加企业的服务成本和管理的复杂程度，所增加的个性化成本和实际收益需要成正比。
图 3 提供个性化旅游服务
携程的大数据应用从用户的角度岀发，分析基于携程所有用户的数据，包括用户在查询、浏览、预订、出行、评论等一系列旅行前后行为中所产生的数据。携程在剔除无效数据的同时，保证用户所留下的数据的真实性，然后将大量的数据进行实时筛选、分拣与重新组织并应用到用户的出行前、出行中、出行后的个性化需求中，如图 3 所示。
要做到个性化，明确用户的目标需求是至关重要的，不仅要看订单，还要关心用户所关心的内容。例如，同样是预订五星级酒店，有些用户对酒店设施十分敏感，有些看重酒店位置，有些则更在意酒店服务，对此，携程会根据用户的需求推荐不同的酒店。
美国塔吉特（Target）百货设立了一个迎婴聚会登记表，并对登记表中顾客的消费数据进行建模分析。他们发现，许多孕妇在第二个妊娠期的开始会买许多大包装的无香味护手霜，在怀孕的最初 20 周会大量购买补充钙、锌之类的保健品。
塔吉特最终选出了 25 种典型商品的消费数据，构建了“怀孕预测指数”。通过这个预测指数，塔吉特能够在很小的误差范围内预测到顾客的怀孕情况，从而就能在合适的时间把孕妇优惠广告寄发给顾客。
“Nike 跑鞋或腕带传感器”使耐克逐渐成为大数据营销的创新公司。运动者只要穿着 Nike 的跑鞋运动，与之关联的 iPod 就可以存储并显示运动日期、时间、距离、热量消耗值等数据。
Nike 通过跑步者上传的跑步路线掌握了主要城市最佳跑步路线的数据库，而且组织城市的跑步活动效果更好。目前，Nike 的运动网上社区有超过 500 万名活跃用户每天不停地上传数据，Nike 借此与消费者建立了前所未有的牢固关系。同时，海量的数据对于 Nike 了解用户习惯、改进产品、精准投放和精准营销也起到了不可替代的作用，Nike 甚至掌握了跑步者最喜欢听的歌是哪些。个性化服务离不开顾客的主动参与和分享，来源于客户的数据也能更精准地服务于客户。
“三只松鼠”近几年的快速发展，一方面是依靠品牌推广，另一方面是在数据分析的基础上不断完善细节，包括个性化的称呼、“三只松鼠”的卡通形象、赠品的差别化、不同的顾客标签分类以及用户体验等。“三只松鼠”通过 ERP 系统能够了解所有顾客在商城的购买记录，通过 CRM 系统能够准确抓取用户的评价，一些不经意的留言和评级会反映出他们的需求。
通过分析顾客过去在商城的购买习惯，用户的购买评价，来判断哪种口味的产品在哪个地区卖得最好，哪种产品是消费者最乐于接受的，从而进行更有针对性的产品首页推荐。同时，他们会对顾客进行个性化、人性化的标签分类和细化分析，从而根据这些分类，推送不同的产品类型。例如，爱老婆型顾客购买的产品主要是以老婆食用为主的，“三只松鼠”会在包裹里放上书信，以“松鼠”的口吻代替顾客给他老婆写一封信。
3. 商品个性化推荐
随着电子商务规模的不断扩大，商品数量和种类快速增长，顾客需要花费大量的时间才能找到自己想买的商品。
个性化推荐系统通过分析用户的行为，包括反馈意见、购买记录和社交数据等，以分析和挖掘顾客与商品之间的相关性，从而发现用户的个性化需求、兴趣等，然后将用户感兴趣的信息、产品推荐给用户。
个性化推荐系统针对用户特点及兴趣爱好进行商品推荐，能有效地提高电子商务系统的服务能力，从而保留客户。
1）电子商务网站
随着电子商务的蓬勃发展，推荐系统在互联网中的优势地位也越来越明显。
在国际方面，Amazon 平台中采用的推荐算法被认为是非常成功的。在国内，比较大型的电子商务平台网站有淘宝网（包括天猫商城）、京东商城、当当网、苏宁易购等。
在这些电子商务平台中，网站提供的商品数量不计其数，网站中的用户规模也非常巨大。据不完全统计，天猫商城中的商品数量已经超过了 4 000 万。
在如此庞大的电商网站中，用户根据自己的购买意图输入关键字查询后，会得到很多相似的结果。用户在这些结果中也很难区分异同，难于选择合适的物品，推荐系统能够根据用户兴趣为用户推荐一些用户感兴趣的商品。电子商务网站利用推荐系统为用户推荐商品，方便了用户，从而也提高了网站的销售额。
2）电影视频网站
个性化推荐系统在电影和视频网站中的应用也很广泛，能够帮助用户在浩瀚的视频库中找到令他们感兴趣的视频。在该领域成功使用推荐系统的一家公司就是 Netflix。
Netflix 原先是一家 DVD 租赁网站，后来开始涉足在线视频业务。Netflix 非常重视个性化推荐技术，并且在 2006 年开始举办著名的 Netflix Prize 推荐系统比赛，希望研究人员能够将 Netflix 的推荐算法的预测准确度提升 10%。
该比赛对推荐系统的发展起到了重要的推动作用：一方面该比赛给学术界提供了一个实际系统中的大规模用户行为数据集（40 万用户对 2 万部电影的上亿条评分记录）；另一方面，在 3 年的比赛中，参赛者提出了很多推荐算法，大大降低了推荐系统的预测误差。
图 4 是 Netflix 的电影推荐界面，包含了电影的标题和海报、用户反馈和推荐理由三部分。Netflix 使用的是基于物品的推荐算法，即给用户推荐和他们曾经喜欢的电影相似的电影。Netflix 宣称有 60% 的用户是通过其推荐系统找到感兴趣的电影和视频的。
图 4 Netflix 电影推荐
YouTube 作为美国最大的视频网站，拥有大量用户上传的视频内容。为了解决视频库的信息过载问题，YouTube 在个性化推荐领域也进行了深入研究，现在使用的也是基于物品的推荐算法。实验证明，YouTube 个性化推荐的点击率是热门视频点击率的两倍。
3）网络电台
个性化网络电台也很适合进行个性化推荐。首先，音乐很多，用户不可能听完所有的音乐再决定自己喜欢听什么，而且每年新的歌曲在以很快的速度增加，因此用户无疑面临着信息过载的问题。其次，人们听音乐时，一般都是把音乐作为一种背景乐来听，很少有人必须听某首特定的歌。对于普通用户来说，听什么歌都可以，只要能够符合他们当时的心情就可以了。因此，个性化音乐网络电台是非常符合个性化推荐技术的产品。
目前有很多知名的个性化音乐网络电台。国际上著名的有 Pandora 和 Last.fm，国内的代表则是豆瓣电台。这 3 个个性化网络电台都不允许用户点歌，而是给用户几种反馈方式：喜欢、不喜欢和跳过。经过用户一定时间的反馈，电台就可以从用户的历史行为中获得用户的兴趣模型，从而使用户的播放列表越来越符合用户对歌曲的兴趣。
Pandora 的算法主要是基于内容的，其音乐家和研究人员亲自听了上万首来自不同歌手的歌，然后对歌曲的不同特性（如旋律、节奏、编曲和歌词等）进行标注，这些标注被称为音乐的基因。然后，Pandora 会根据专家标注的基因计算歌曲的相似度，并给用户推荐和他之前喜欢的音乐在基因上相似的其他音乐。
Last.fm 记录了所有用户的听歌记录及用户对歌曲的反馈，在这一基础上计算出不同用户在歌曲上的喜好相似度，从而给用户推荐和他有相似听歌爱好的其他用户喜欢的歌曲。同时，Last.fm 也建立了一个社交网络，来让用户能够和其他用户建立联系，以及让用户给好友推荐自己喜欢的歌曲。Last.fm 没有使用专家标注，而是主要利用用户行为计算歌曲的相似度。
4）社交网络
社交网络中的个性化推荐技术主要应用在 3 个方面：利用用户的社交网络信息对用户进行个性化的物品推荐，信息流的会话推荐和给用户推荐好友。
Facebook 保存着两类最宝贵的数据：一类是用户之间的社交网络关系，另一类是用户的偏好信息。
Facebook 推出了一个称为 Instant Personalization 的推荐 API，它能根据用户好友喜欢的信息，给用户推荐他们的好友最喜欢的物品。很多网站都使用了 Facebook 的推荐 API 来实现网站的个性化。
著名的电视剧推荐网站 Clicker 使用 Instant Personalization 给用户进行个性化视频推荐。Clicker 现在可以利用 Facebook 的用户行为数据来提供个性化的、用户可能感兴趣的内容“’流”了，而更重要的是，用户无须在 Clicker 网站上输入太多数据（通过评分、评论或观看 Clicker.com 上的视频等方式），Clicker 就能提供这样的服务。
除了利用用户在社交网站的社交网络信息给用户推荐本站的各种物品外，社交网站本身也会利用社交网络给用户推荐其他用户在社交网站的会话。每个用户在 Facebook 的个人首页都能看到好友的各种分享，并且能对这些分享进行评论。每个分享和它的所有评论被称为一个会话，Facebook 开发了 EdgeRank 算法对这些会话排序，使用户能够尽量看到熟悉的好友的最新会话。
除了根据用户的社交网络及用户行为给用户推荐内容，社交网站还通过个性化推荐服务给用户推荐好友。
5）其他应用
Because e-commerce enterprises has basically realized the data of all aspects of the business process, so you can take full advantage of big data mining techniques to analyze these data to optimize their business processes and increase business profits. In addition to the previously described several applications, e-commerce industry in large data can also be used in many other ways.
① dynamic pricing and special offers
E-commerce companies can use the data to build customer data, and found that users like how much to spend and like what products to buy, so by tracking customer behavior, use big data analytics to develop flexible pricing and discount policies. For example, if your analysis shows user interest in a particular category of goods soared, the e-commerce companies can offer a discount or buy one get one offer.
② customized deals
E-commerce businesses can determine customer buying habits by using the data and send them to targeted special offers and discount codes based on previous buying patterns. Data can also be used to re-suspend attract customers when the customer purchases or just do not buy, for example, to remind customers that they viewed product or by sending an e-mail inviting them to complete the purchase.
③ Supply Chain Management
E-commerce companies can use big data to more effectively manage the supply chain. Data analysis can reveal any potential inventory problems or delays in the supply chain. If an item there is a problem, it can be immediately removed from the sale, so as not to damage customer service issues.
④ predictive analysis
Predictive analysis refers to a variety of channels using a large e-commerce business data analysis techniques to help companies develop business plans for future operations. Data analysis may show new buying trends online store business enterprise sector slowed or sales of goods.
We can use this information to help plan the next phase of the inventory, and to develop new target market. Keep abreast of the latest trends in e-commerce has some challenges, but the use of Big Data technologies can greatly enhance the profits of enterprises and help enterprises to establish a successful forward-thinking business. If you do not use the power of big data mining, it may miss the opportunity to market success.
Recommended Learning Catalog: application of the Internet industry in large data