The traditional sorting model mainly includes relevance and importance: bool model, VSM, language model importance: PageRank, trustrank
1. bool Model
Query is a logical expression, that is, "and/or/Not". similarity is determined by Boolean algebra. Only correlation is irrelevant.
2. VSM
It is an algebraic model that represents a document. The ing of the document is the T-dimension feature vector, and the weight of each dimension feature mainly has multiple variants such as TF-IDF. Similarity Calculation mainly refers to cosine Similarity
3. probability search model
Bm25 splits the query into multiple elements, and considers the binary independence model (equivalent to IDF) of each element, the weight value of the element in the document, the weight value of the element in the query, and the sum.
4. Language ModelCreate different language models for each document, determine the likelihood of query generation by the document, and then sort the generated probability from high to low as the search result. Data sparse problem: Many query words are not found in the document, resulting in a probability of 0. Retrieval failure solution: Use the background probability for data smoothing, the background probability of a single occurrence is the number of occurrences divided by the total number of occurrences of a single occurrence in the document set, the formula for calculating the query probability generated by a document is the document language model of each query word + the smooth document set language model, and the product is obtained.
Improvement: Hmm, relevant model, and translation model
5. PageRank
