TMDB Score shaping

Get Started. It's Free
or sign up with your email address
TMDB Score shaping by Mind Map: TMDB Score shaping

1. Untitled

2. signal modeling

2.1. analysis process

2.1.1. Untitled

2.2. ultimate goal of analysis is striking a balance between precision and recall

2.2.1. Untitled

2.3. how to improve recall

2.3.1. stemming Untitled

2.3.2. synonyms specificity capturing meaning acronyms Untitled

2.3.3. edge grams Untitled

2.4. how to improve precision

2.4.1. token filter to remove stop words such as OR, AND Untitled

2.4.2. shingles Untitled

3. ranking func design

3.1. multifield search

3.1.1. multi-match best_fields tie_breaker most_fields coord cross_fields

3.1.2. match_phrase

3.2. field-centric / term-centric

3.2.1. default scoring DF * ITF Untitled

3.2.2. problem search multiple terms in multiple fields and give out an score

3.2.3. two basic solutions Untitled

3.2.4. field-centric search two forms best_fields most_fields baked into multi_match best_fields most_fields

3.2.5. field-centric problems albino elephant def sample albino elephant albino elephant in start trek signal discordance def star trek example

3.2.6. term-centric search dismax-style Untitled benefits downsides fine-tune solve signal discordance custom all_fields cross_fields cross_fields vs custom all_fields

3.2.7. combine field/term-centric search like fields together group "like fields" together limits of like fields combine greedy naive search / conservative amplifiers two factors Untitled

3.3. shaping the relevance function

3.3.1. functional query

3.3.2. boolean query via Boolean clause

3.3.3. example problem Untitled reason TF*IDF has a strong bias towards shorter fields through field normalization solution additive boosting multiplicative boosting filtering

3.4. movie search ranking rules

3.4.1. scoring layers Untitled

3.4.2. Untitled exact name matching def boolean boost on exact-title matching add a clause for big rammed matches another two layers Untitled final complete query Untitled