Big Data: Web ads

1. targeted advertising: Web advertisements are selected based on the interests of a user, so that web services can run with AD benefits.

2. online and offline algorithms: the traditional algorithms that produce answers only when all data is obtained are called offline algorithms. Online algorithms must answer every element in the stream immediately. At this time, they only know the information of the past and have no idea about the future data.

3. Greedy Algorithm: Online algorithms adopt greedy policies. each step of the algorithm is selected based on the minimization of a certain target function.

4. Competition rate: in all possible input cases, the quality of online algorithms is measured by minimizing the benefits ratio between online algorithms and optimal offline algorithms.

5. Two-part Graph Matching: Two node sets, which are connected by two collection nodes to form a set of edges, maximizing the number of edges. Each node will not appear more than once.

6. Online solution to the matching problem: Find a greedy algorithm in the two diagram to sort the edges in a certain way and process each edge in sequence. It can be proved that the competition rate of this algorithm is 1/2.

7. Search Advertisement Management: Search Engines receive bids from advertisers for certain search queries. For a search query, an advertisement is displayed. Once an advertisement is clicked, The Advertiser will pay for it.

8. adwords: The adwords data includes a series of bidding sets by advertisers for certain search queries, the total budget of each advertiser, and the historical click rate of each ad after each query is submitted. There is also a search query stream received by the search engine. The goal is to display a fixed online ad set for each query to maximize the search engine benefits.

9. Simplified adwords: Each bidding price is not 0 or 1. Each query corresponds to only one advertisement, and the budget of all advertisers is equal. Greedy algorithms enable search engines to distribute ads to advertisers who have bids for queries and budget remaining. Competition rate 1 is 1/2.

10 Balance Algorithm: Compared with greedy algorithms, the balance algorithm will allocate the ads to the advertisers who have the most budget for the query. Once the remaining budget of multiple advertisers is equal, you can select them randomly. The competition rate between two advertisers is 3/4, and that between advertisers is 1-1/E, which is about 63%.

11. The balance algorithm for general adwords problems: advertisers have different prices, different budgets, and different query Ctr. The balance algorithm distributes ads to the highest function x * (1-power (-f )) value of the advertisers. X: the product of the bidding price and click rate. F is the budget comparison column not used by advertisers.

12 adwords implementation: the bidding keywords are consistent with the query. The query is represented as a sort table of words. The bidding is saved in a hash table or similar structure, and the hash key is the sort table.

13 word set and document match: If all words in the bidding set appear in the document, whether or not the words are in the same order as the bidding set, or whether they are adjacent or not, the bidding and document match are considered.

Hash Storage of the 14-word set: store the words in the bidding set according to the low frequency priority, and use the first word as the hash key.

15 document handling in bidding matching: the words in the document are prioritized by low frequency.