In the past two years, artificial intelligence has accelerated and been applied to all walks of life. So, in the field of hitchhiker, with the use of more artificial intelligence and advanced algorithms, what new possibilities have been brought to the renewal of experience efficiency?
Recently, Dida Travel has launched two new functions of order prediction that are the first in the tailwind industry: "Estimated Response Time" and "Recommended Departure Time". For instant departure orders, Dida Travel is based on a newly developed AI algorithm model to help hitchhiker users accurately estimate that a car owner will take the order in a few minutes. At the same time, based on the estimation of the probability of receiving orders from car owners at different points in time, it recommends departure time points with higher probability of receiving orders for users, so as to further improve the certainty of travel.
In order to give readers a deeper understanding of the scientific principles behind the rejuvenation of the ride-up experience empowered by artificial intelligence and advanced algorithms, Dida Travel today launched the "Tick Cool Technology. The theme of the fourth phase of the "Science Experience Hall" is how AI models can bring deterministic new experiences to rideshare users based on accurate predictions.
In the popular science experience hall, through the deconstruction of the birth process of the AI model, Dida Travel tells in simple terms how the AI model can accurately predict the estimated response time step by step. The relevant person in charge of Tick Travel said that hitchhiking is different from commercial travel, and the owners and passengers have their own travel needs, and they are voluntary and independent carpooling, so how to make users have a sense of control over hitchhiking and reduce anxiety about uncertainty through intelligent technology is an important direction for hitchhiking experience upgrading. "We hope that through more cutting-edge intelligent technologies, users can make every ride-sharing more efficient, faster and worry-free, and better meet the growing needs of users for instant travel."
In recent years, Tick Travel has continued to consolidate its technical infrastructure, accelerated the upgrading of artificial intelligence, advanced algorithms, cloud technology, and navigation capabilities, effectively solved many pain points that have long restricted the efficiency of the ride-hailing experience, and also promoted the continuous improvement of the ride-hailing response rate, with the tick-ride response rate of 66.5% in 2023, significantly higher than the industry average. At the same time, Tick's proprietary technology supports approximately 45,000 route planning requests per second, optimizing the nearest pick-up location and minimizing waiting times and travel distances.
The immediacy of hitchhiking is further improved, and the AI model further improves the certainty of the trip
In the field of mobility, AI algorithms have long been used for efficient and real-time large-scale supply and demand matching scheduling. In the past two years, with the breakthrough of artificial intelligence, as well as the development of neural networks and deep learning, algorithm models have been able to solve more complex and high-order problems, and the accuracy and efficiency have been further improved.
For example, in the past, when a hitchhiker placed an order and waited for the order to be received, the platform would generally inform the passenger, "It usually takes 10 to 15 minutes for the car owner to take the order, please be patient." Nowadays, after the hitchhiker places an order, it can be accurately informed, and the owner is expected to take the order within a few minutes.
For the new function of "recommended departure time", when passengers want to depart immediately, the platform will automatically recommend a time point with a higher probability of receiving orders to passengers around the passengers' departure time to ensure smooth carpooling. "In fact, a lot of AI-enabled experience improvements are behind the scenes, making users feel that everything is natural and more comfortable and enjoyable in every aspect of travel." ”
According to the big data of Tick Hitch, according to the statistics of the next order, the proportion of hitchhiking orders departing in the next 30 minutes has exceeded 4%, and the proportion of orders departing in the next 15 minutes is close to 7%. At the same time, the average response time for orders departing within 30 minutes has been shortened to more than 1.6 minutes, and the average response time for orders departing within 15 minutes has been less than 1.3 minutes. This shows that more and more car owners and passengers are taking hitchhiking as an important choice for instant travel, placing orders and receiving orders in real time, and setting off immediately.
The big data-based matching, scheduling algorithm and navigation not only effectively improve the certainty of ride-hailing, but also continuously improve the response rate of tick-hitchhiking. According to the data, in 2021, 2022 and 2023, the response rate of Tick Hitch will be 56.4%, 58.8% and 66.5% respectively, which is significantly higher than the industry average.
The in-depth application of AI models makes the pre-ride prediction more comprehensive and accurate
With the continuous upgrading of big data, intelligent strategies and navigation capabilities, the pre-departure prediction range of Tick Hitch is getting wider and higher, and the accuracy is getting higher and higher, which further improves the certainty of hitchhiking. In this popular science experience hall, Tick Travel took the estimated response time model as an example to show the birth process of AI small models in the field of travel in simple terms.
The first step is to gain insight into user needs. Riders want to know that the owner will take the order a few minutes after placing the order, so as to increase the certainty of the trip. For those who are in a hurry to set off and travel across cities, this demand will be even stronger.
The second step is data collection and processing. The estimated response time model needs to refer to the average response time of car owners at each point in time, the response time of different types of orders, the historical supply and demand of the route and the number of car owners for integrated analysis in the past period of time, so as to enter the third step of feature extraction, and extract differentiated features by analyzing the characteristics of these data samples.
Step 4: Model Selection. According to the characteristics of the data and the specific problem to be solved, different algorithms are selected, such as linear regression, decision tree, neural network, logistic regression, large model, etc.
The fifth step is model training and tuning. Based on offline data, the model is continuously trained, the prediction accuracy is evaluated, and the optimization is continuous, and when the standard is reached, it is launched online, and it continues to evolve iteratively in practice. Therefore, each model has been honed before it is officially launched.
The sixth step is to export the predicted recommended score. Rideshare passengers can see in real time after placing an order, and it is expected that a car owner will take the order in a few minutes.
The above six steps are also the general birth logic of the mobility AI model, but they are different in terms of specific requirements, data dimensions, data characteristics, model selection, and the type of exported results. For example, for the estimated order probability model, the characteristics used include the order time, route, number of people, exclusive or combined, weather, the number of car owners on the route, and historical orders.
Tick Travel algorithm engineer said: In the process of the birth of a new AI model, there is a lot of room for innovation in each step of sample selection, feature extraction, and model structure design optimization, and more in-depth insights into user needs, higher quality data, more accurate feature extraction, and more appropriate model selection and tuning, all of which can make the final export results more accurate.
The field of mobility requires more proprietary models, and each model is a scientific experiment
Of course, accurate prediction is only one of the innovative values of AI models in the field of hitch travel, and they are becoming more and more accurate and efficient with the application of big data and advanced algorithms in terms of starting and ending point search, pick-up point recommendation, path planning, drop-in matching, intent recognition, reasoning and analysis, etc., allowing machines to replace humans to complete more complex work in more fields and links.
The implementation of these functions is actually based on small AI models. Tick Travel algorithm engineer said that at present, Tick Travel has started the application of neural networks and deep learning models, which are characterized by more complex network structure, more feature crossing, more parameters, and more accurate, real-time and efficient derivation results or specific actions.
In response to the application logic of artificial intelligence in the field of mobility, the person in charge of the intelligent strategy of Tick Travel said, "There are many links in travel, and it cannot be defined by a complete and unified problem. We need to break down the finer granularity of the problem to solve, and analyze how each link can improve the experience. Each model solves different problems and has different training goals. The logical path to solve the pain point is still machine learning, which goes to deep learning from a large number of samples and continuously optimizes the model in practical training. ”
In the short term, the field of mobility does not need a general large model, but should take a dedicated route, through different proprietary models to perform their own duties, to solve specific problems in specific business scenarios, and finally the more data, the more accurate the model export results, the higher the degree of automation, and the more intelligent the travel app.
The process of using artificial intelligence and advanced algorithms to solve the pain points of hitchhiking experience efficiency is itself an innovative experiment from 0 to 1, based on which more new algorithm models will be born, allowing hitchhiking to enter a new era of intelligence.