The formula for Sphinx default ranker, SPH_RANK_PROXIMITY_BM25 looks like this:
SPH_RANK_PROXIMITY_BM25 = sum(lcs*user_weight)*1000+bm25
The Longest Common Subsequence is computed for each field separately and then multiplied by user_weight. However bm25 is just a document-wide variable and does not take user fields into account. Why is that so?
Just because it's faster and in many cases the quality is enough. There's a custom ranker and bm25f to be used there. Document length is also not accounted by default, it requires index_field_lengths=1 during indexing.
Related
I have a list of chemicals in my database and I provide our users with the ability to do a live search via our website. I use SQLAlchemy and the query I use looks something like this:
Compound.query.filter(Compound.name.ilike(f'%{name}%')).limit(50).all()
When someone searches for toluene, for example, they don't get the result they're looking for because there are many chemicals that have the word toluene in them, such as:
2, 4 Dinitrotoluene
2-Chloroethyl-p-toluenesulfonate
4-Bromotoluene
6-Amino-m-toluenesulfonic acid
a,2,4-trichlorotoluene
a,o-Dichlorotoluene
a-Bromtoluene
etc...
I realize I could increase my limit but I feel like 50 is more than enough. Or, I could change the ilike(f'%{name}%')) to something like ilike(f'{name}%')) but our business requirements don't want this. What I'd rather do is improve the ability for Postgres to return results so that toluene is at the top of the search results.
Any ideas on how Postgres' ilike capability?
Thanks in advance.
One option is to better rank the results. Postgres text search allows you to rank the results.
A cheap and dirty version of preferential ranking is to do multiple queries for name = ?, ilike(f'{name}%')), and ilike(f'%{name}%')) using a union. That way the ilike(f'{name}%')) results come first.
And rather than a hard limit, offer pagination. SQLAlchemy has paginate to help.
ILIKE yields a boolean. It doesn't specify what order to return the results, just whether to return them at all (you can order by a boolean, but if you only return trues there is nothing left to order by). So by the time you are done improving it, it would no longer be ILIKE at all but something else completely.
You might be looking for something like <-> from pg_trgm, which provides a distance score which can be sorted on. Although really, you could just order the result based on the length of the compound name, and return the shortest 50 that contain the target.
something like ilike(f'{name}%')) but our business requirements don't want this
Isn't your business requirement to get better results?
But at least in my database, this could just return a bunch of names in inverted format, like toluene, 2,4-dinitro, so the results might not be much better, unless you avoid storing such inverted names. Sorting by either <-> or by length would overcome that problem. But they would also penalize toluene, ACS reagent grade 99.99% by HPLC, should you have names like that.
I have a list of titles and descriptions in a table which are indexed in a tsvector column. How can I implement Google Search like full text search functionality in Postgres for these fields. I tried various functions offered by standard Postgres like
to_tsquery('apple | orange') -- apple | orange
This function returns rows as long as it has one of these terms so it doesn't produce highly relevant results at top which should have both of the terms.
plainto_tsquery('apple orange') -- apple & orange
This function requires all of the terms in the query. But I want results including both apple and orange first but can still have results including even one of these terms just later in the results.
phraseto_tsquery('apple orange') -- apple <> orange
This function only matches orange followed by apple but not vice versa. But for me orange <> apple is also still relevant.
I also tried websearch_to_tsquery() but it behaves very similar to above functions.
How can I ask Postgres to list highly relevant rows first which contains most of the terms in the search query no matter the order of the terms and then followed by rows with less number of terms?
to_tsquery('apple | orange') -- apple | orange
This function returns rows as long as it has one of these terms so it doesn't produce highly relevant results at top which should have both of the terms.
Unless you tell it how to order the rows, rows of a single query are returned in arbitrary order. There is no "top" without an ORDER BY, there is just something which happens to be seen first.
How can I ask Postgres to list highly relevant rows first which contains most of the terms in the search query no matter the order of the terms and then followed by rows with less number of terms?
Use the | operator, then rank those rows using ts_rank, ts_rank_cd, or a custom ranking function you write yourself. For performance, you might want to use the & operator first, then revert to | if you don't get enough rows.
The built in ranking functions don't care about order, but also don't care about proximity. So they might not do what you want. But writing your own won't be particularly easy, so I'd at least try them out first.
It would be nice if the introduction of websearch_to_tsquery or phraseto_tsquery had also introduced some corresponding ranking functions. But since they invented only ordered proximity, not proximity without order, it is unlikely they would do you want if they did exist.
I have two tables in which I have data coming from two different sources. One of the field of each table contains the title of a movie, but for some reason out of my control, the titles are not always exactly the same.
So I use the ts_vector to get rid of all the minor differences (stop words, plurals and so on).
See an example here: http://sqlfiddle.com/#!17/5ccbc/3
My problem is how to compare the two ts_vector without taking into account the numberic values, but just the text content. If I compare directly the two fields, I only get the exact match between values, including position of each word. The only solution I have found is using the strip() function, that remove positions and weights from tsvector, leaving only the text content.
I was wondering if there is a fastest way to compare ts_vectors.
You could create in index on the stripped vector:
create index on tbl1 (strip(ts_title));
create index on tbl2 (strip(ts_title));
But given that your query has to fetch every row of each table, it is unlikely this would serve much of a point. Doing a merge join between the precomputed stripped vectors could be faster, but probably not once you include the overhead of building and maintaining the indexes. If the real WHERE clause is more restrictive (selecting only a few rows from one or the other of the tables) then please share the real query.
I have a table with 100+ values corresponding to each row, so I'm exploring different ways to store them.
Without any indexes, would I lose anything if I store these 100 values in an integer[] column in postgresql? As compared to storing them in separate columns.
Plus, since we can add indexes to array elemnets,
CREATE INDEX test_index on test ((foo[1]));
Would there be a performance difference queries using such an index as compared to regular index on a column?
As far as I've read, this performance difference would come into picture in arrays with variable length elements; but I'm not sure about fixed length ones.
Don't go for the lazy way.
If you need to store 100 and more values as array, it is ok, if it has sense has array for your application, your data.
If you need to query for a specific element of the array, then this design is not good, regardless of performances, and you must use columns. This will help you in the moment you must delete a "column" in the middle or redesign it.
Anyway, as wrote by Frank in comments, if values are all same type, consider to model them to another table (if also the meaning is the same).
I'm trying to calculate a ts_rank for a full-text match where some of the terms in the query may not be in the ts_vector against which it is being matched. I would like the rank to be higher in a match where more words match. Seems pretty simple?
Because not all of the terms have to match, I have to | the operands, to give a query such as to_tsquery('one|two|three') (if it was &, all would have to match).
The problem is, the rank value seems to be the same no matter how many words match. In other words, it's maxing rather than multiplying the clauses.
select ts_rank('one two three'::tsvector, to_tsquery('one')); gives 0.0607927.
select ts_rank('one two three'::tsvector, to_tsquery('one|two|three|four'));
gives the expected lower value of 0.0455945 because 'four' is not the vector.
But select ts_rank('one two three'::tsvector, to_tsquery('one|two'));
gives 0.0607927 and likewise
select ts_rank('one two three'::tsvector, to_tsquery('one|two|three'));
gives 0.0607927
I would like the result of ts_rank to be higher if more terms match.
Possible?
To counter one possible response: I cannot calculate all possible subsequences of the search query as intersections and then union them all in a query because I am going to be working with large queries. I'm sure there are plenty of arguments against this anyway!
Edit: I'm aware of ts_rank_cd but it does not solve the above problem.
Use the smlar extension (linux only AFAIK, written by the same guys that brought us text search).
It has functions for calculating TFIDF, cosine, or overlap similarity between arrays. It supports indexing so is fast.
Another way would be to "spell-check" the query prior to using it, basically removing any query terms that are not in your corpus.
The conclusion that I have come to is to & the items together for the ranking. In my select query (with which I'm doing the search) the items are |ed. This seems to work.