Team: E-commerce
Position: Machine Learning Engineer - E-commerce
Years: 2025
Interviewee: New Grad
Round: One
Interviewer: Team Lead
Questions.
1. Merge two sorted linked lists(force deducts easy original question in five minutes and seconds)
2. Merge k sorted linked list(force buckle hard original question 5 minutes) and then asked follow up, there is no other way to write, mentioned the practice of priority queue and merge sort, asked to write a, but relatively simple time is very full have written, plus complexity analysis
3. finally there is a follow up, how to de-emphasize, (force buckle original question medium) side merge side de-emphasis or merge finish de-emphasis, 40 minutes to write a bunch of follow up but compared to one side of the three-dimensional dp is still very good to write!
Round: Two
Interviewer: Tech Lead (North American Tiktok interviewer)
Questions.
(1) Self-introduction
(2) Describe your understanding of collaborative filtering, and what scenarios it is used in.
(3) Explain how user-based or item-based calculates similarity. how do you consider rating?
(4) Describe the two-phase recommendation system
(5) What model is used by Candidate generator? Explain the model, model time complexity.
(6) What model does Ranker use, the explanatory model and model complexity
(7) Why use multiple generators
(8) What if we ended up recommending 5 products to our users. Why 5? How do you get these five using a two-phase recommendation system? If you use different labels, explain how the loss functions are calculated for each label.
(9) Scenario: If we have new users or new products coming in, how can you help them get up and running in the system quickly.
(10) If we now have item-embedding and user-embedding, how do you define them? How can they be computed? What model to use? The interviewer firstly called me to explain matrix factorization + loss func + how to optimize; secondly called me to explain what two-tower neural network is.
(11) A med coding question was given at the end.
Result: Passed
Round: Three
Interviewer: Tech lead (Interviewer, Tiktok China)
Questions:
(1) Self-introduction
(2) Explain all the models that have been used in the chatbot project.
(3) Explain why LSTM is used, the framework of LSTM and explain the mathematical modeling in each Gate.
(4) Explain the activation function in neural network : sigmoid, Than, Relu and Leaky Relu. write their functions and describe the advantages and disadvantages.
(5) How it was resolvedReLU DyingProblems and why they occur.Gradient vanishing/exploding solution.
(6) Write the mathematical analytical formula for back propagation
(7) Write the loss function of a sigmoid function and write in the form of a formula how to optimize to find the optimal parameters. both methods of maximum likelihood and entropy are required to be explained and formulated.
(8) A complete description of a recommender system system design (with an intermediate focus on asking how the FEATURES section was obtained and designed)
(9) A med code.
Round: Four
Interviewer: Recruiter
Questions:
The final interview is with HR, the interviewer mainly wants to know more about your background, skills, career plan, and assess whether you fit the company culture and development direction.
1. Self-introduction
2. Why TikTok
3. Biggest challenges in the workplace
4. Advantages and disadvantages
5. How to deal with disagreements with others
6. Career planning
With our interview assistance, the candidate successfully passed the interview for Tiktok, if you also need our interview assistance service, please contact us now!Contact Us.
