https://mp.weixin.qq.com/s/PIzxQE8Q2HLSYtr13FoaMQ
By 超神经
场景描述：将 AI 技术如计算机视觉、语音识别、自然语言处理、大数据分析等应用到旅游服务业中，一方面可帮助游客减少不必要的时间与金钱浪费，另一方面，可提升景区服务效率与质量，实现双赢局面。
关键词：爬虫 计算机视觉 语音识别 自然语言处理 大数据分析 云计算The "May Day" holiday is over. Have you calculated where your vacation time is spent? Traffic jam? Line up for attractions? Still waiting for a C person to take pictures?
It is said that this four-day holiday gave the majority of tourists another Spring Festival travel experience. It is difficult to find train tickets and tickets for popular scenic spots. The scalpers took the opportunity to make a lot of money:
During the "May Day" period, Shanghai Disneyland was too popular,
causing the ticket price to be doubled as usual
Someone joked that after returning from a May Day holiday, he finally understood Thanos’ feelings.
Data analysis helps tourists avoid crowds
Faced with this situation, the savvy programmer thought of a way to help everyone wisely avoid the crowds. |Source: Programmers grow together
Based on the data obtained through crawlers, the little brother of the programmer found that the most popular attraction on May 1 this year was Shanghai Disneyland, so he decided not to join in the fun.
After analyzing the data of popular attractions in some cities, get the popularity ranking, and see if there is any contribution from these data.
The sales ranking results of the above attractions are obtained by analyzing the sales of tickets in popular tourist cities on the Qunar website. The specific method is as follows:
By requesting https://piao.qunar.com/ticket/list.htm?keyword=Beijing, get the information of popular scenic spots in Beijing, and then analyze and extract the required information through BeautifulSoup.
According to the climbed data, you can not only analyze the ticket sales of each scenic spot, but also make a heat map of the scenic spot. Take Beijing as an example:
It can be seen from the heat map that during the "May Day" period this year, the Forbidden City Museum, Prince Gong's Mansion, Beijing Zoo and other attractions are the most popular
The open platform of Baidu Maps is used here. First, you need to register the developer information. There is a "Apply Secret Key" button at the bottom of the homepage, just click to create it.
Since the application type is selected on the browser side, you only need to assemble the data and replace the corresponding html code. In addition, you need to replace the AK that you access the application.
Finally, the data was used to compare the prices of popular scenic spots in order to make corresponding plans according to their own budget.
Another programmer used Python to crawl the scenic spots data of all provinces and autonomous regions in the country, and got the heat map of national scenic spots during the National Day last year:
From this figure, it can be clearly seen that the heating value of Beijing-Tianjin-Hebei, Yangtze River Delta, Pearl River Delta and Sichuan ranks the top four
With these analyses, you can make a reasonable plan for the next vacation to avoid these hot spots, and you don't have to worry about being stuck on the road.
AI helps scenic spots improve efficiency
In addition to tourists, during long and small holidays, scenic spots and gourmet shops are also troubled by people.
During the "May Day" period, many scenic spots were blocked and exploded. The Badaling Great Wall welcomed 54,000 tourists on the first day of the holiday, and generated nearly 200,000 tons of garbage that day. The just-opening Expo welcomed 327,000 visitors, and the spectacular scenes can be imagined.
Tourists on the Great Wall fully experienced what is called a "dilemma"
An online celebrity lobster restaurant in Changsha issued a letter of apology, saying that it was unable to receive many guests, and on the list of numbers it displayed, there were 7172 tables.
Faced with these problems, many companies are also actively using artificial intelligence to help scenic spots find solutions.
The earliest AI + tourism is cut from a single scene in order to promote the intelligence of the tourism industry. For example, audio guides/translations, robot customer service, AR/VR digital guides, etc. that we will come into contact with in many attractions. But these applications are all independent of each other, and there is still no fundamental improvement for the tourist experience.
In recent years, some domestic companies have taken a new step and began to provide a full range of operational analysis basis for scenic spots and local tourism departments.
Aibee's first batch of projects ``Swipe the face into the park'' has been put into use in Kanas, Qinhuang Terracotta Warriors and Horses, Gubei Water Town and other scenic spots
For example, Aibee, founded by Dr. Lin Yuanqing, the former director of Baidu's deep learning laboratory, combines AI with offline travel to provide solutions for global travel.
They open up online and offline data. First, tourists can register to buy tickets by swiping their faces. After entering the park, they will track the tour of tourists. During the tour, the scenic spot can use the form of recommendation, pop-up window, etc. to push information and service to find people, and then promote the second consumption of tourists inside and outside the scenic spot.
Aibee’s solution involves multi-modal AI technology including computer vision, speech recognition, natural language understanding, big data analysis, etc. The technology is not complicated, but combining these technologies to propose an overall solution is The improvement of service efficiency and quality of scenic spots has provided great help.
At present, Aibee has cooperated with 65 scenic spots such as Wudang Mountain, Kanas, and Huashan. Its face recognition system only takes a few seconds, while the traditional ticket checking method takes at least 30 seconds. The efficiency of entering the park has increased by about 10 times.
BAT's AI + tourism layout
Facing the big market of tourism, how can BAT just be the audience?
Tencent: In August last year, Tencent and Chongqing jointly created "A Mobile Tour of Wulong", involving the Internet of Things, cloud computing, big data, artificial intelligence and other technologies, through personalized line customization, expert recommendation, intelligent customization, intelligent navigation, etc. Function to provide tourists with one-stop intelligent services in tourism information acquisition, itinerary planning, product booking, travel notes sharing, and special e-commerce purchases.
Ali: Last year, Ali’s first physical store of the future hotel "FlyZoo Hotel" (Chinese name "FlyZoo Hotel") opened. It is said that this is the first hotel in the world that supports full-scene facial accommodation. From check-in to guest rooms The experience and check-out process realizes AI intelligent services throughout the whole process.
The robots in the "Feizhubuthi" hotel are delivering meals
Baidu: For tourism authorities and scenic spots, Baidu also actively leverages its data and technical advantages such as search, portraits, public opinion, knowledge graphs, and artificial intelligence, to fully cover tourists before, during and after the trip, and comprehensively deploy tourism big data , Provide visual platform (SaaS) and data interface (API) services to help scenic spots improve precision marketing capabilities, optimize safety management efficiency, improve tourist service quality, and guide global tourism layout.
Baidu's AI + travel solution architecture
Therefore, before you go out on vacation next time, you may wish to use the power of technology to make a thermal analysis of the scenic spots in the list, avoid the crowds, and calmly cross the mountains and the sea.
HyperNeuropedia
Multimodal Learning
Modality refers to the specific way people receive information. Since multimedia data is often the delivery medium of multiple information (for example, textual information, visual information and auditory information are often disseminated in a video), multi-modal learning has gradually developed into the main means of multimedia content analysis and understanding.
Multimodal learning mainly includes the following research directions:
Multi-modal representation learning: It mainly studies how to numerically convert the semantic information contained in multiple modal data into real-valued vectors.
Inter-modal mapping: The main research is how to map the information in a particular modal data to another modal.
Alignment: Mainly research how to identify the correspondence between parts and elements between different modalities.
Fusion: Mainly research how to integrate models and features between different modalities.
Collaborative learning: The main research is how to transfer the knowledge learned in the information-rich modal to the information-poor modal, so that the learning of each modal can assist each other. Typical methods include multi-modal zero-sample learning and domain adaptation.
