Sphinx ignores ranking, always sorts in the same order

问题

I have sphinx index on schools, and when I do a query I always receive the same results, in the same order. I've tried every imaginable combination of ranking, sorting, and matching, and always get the same sorting.

A sample of the bad data I'm getting is below:

"albany high"

Albany Junior High School | Auckland, NZ | 2001 (shouldn't be first)
Albany High School        | Albany, NY   | 2001
South Albany High School  | Albany, OR   | 2001
Albany High School        | Albany, CA   | 1001 (shouldn't be last)

As you can see, the highest-ranked school is not in a city named "Albany", and should be lower, while the lowest-ranked "Albany High School" should be ranked higher than it is. This problem is replicated across many search terms.

The Sphinx index looks like this:

source schools : root
{
    sql_query = \
        SELECT schools.id, schools.name, schools.state, schools.country, schools.city, \
        (select COUNT(*) from user2school WHERE school_id = schools.id) as user_count \
        FROM schools

    sql_attr_uint       = user_count
}

index schools
{
    source                  = schools
    path                    = /var/db/sphinx/data/schools
    min_infix_len           = 3
    infix_fields            = name
}

The code that generates the results is as follows:

$sphinx->SetMatchMode(SPH_MATCH_EXTENDED);
$sphinx->SetRankingMode(SPH_RANK_WORDCOUNT);
$sphinx->SetSortMode(SPH_SORT_RELEVANCE);

$sphinx->SetFieldWeights(array(
    'id' => 0,
    'name' => 1000,
    'city' => 0,
    'state' => 0,
    'user_count' => 0
));

How can I get Sphinx to recognize my custom weights? Every combination I've tried seems to fail.

Edit:

Here is another example with the same ordering, but totally different settings. The only option I have turned on here is:

$sphinx->SetRankingMode(SPH_RANK_SPH04);

The results:

"albany high"

Albany Junior High School | Auckland, NZ | 3 (still shouldn't be first)
Albany High School        | Albany, NY   | 3
South Albany High School  | Albany, OR   | 2
Albany High School        | Albany, CA   | 1 (still shouldn't be last)

As you can see, the ordering is identical. It is identical in every combination of ranking, sorting, and weighting I have tried. Is there anything I can try to debug this problem?

回答1:

Perhaps its a logic error in your application. Sphinx gives you a list of IDs, which you would then use to retreive data from the original database. Maybe you arent sorting those rows right.

I just tried inserting your data into a test RT index (including a string attribute, so could see the data)

mysql> insert into rttest values (1,'Albany Junior High School','Auckland','NZ','Albany Junior High School, Auckland, NZ');
   ... etc ...

mysql> select * from rttest where match('albany high');
+------+--------+-----------------------------------------+
| id   | weight | value                                   |
+------+--------+-----------------------------------------+
|    2 |   3267 | Albany High School, Albany, NY          |
|    3 |   3267 | South Albany High School, Albany, OR    |
|    4 |   3267 | Albany High School, Albany, CA          |
|    1 |   1304 | Albany Junior High School, Auckland, NZ |
+------+--------+-----------------------------------------+
4 rows in set (0.15 sec)

mysql> select * from rttest where match('albany high') option ranker=sph04;
+------+--------+-----------------------------------------+
| id   | weight | value                                   |
+------+--------+-----------------------------------------+
|    2 |  12267 | Albany High School, Albany, NY          |
|    4 |  12267 | Albany High School, Albany, CA          |
|    3 |  10267 | South Albany High School, Albany, OR    |
|    1 |   6304 | Albany Junior High School, Auckland, NZ |
+------+--------+-----------------------------------------+
4 rows in set (0.00 sec)

mysql> select * from rttest where match('albany high') option ranker=wordcount;
+------+--------+-----------------------------------------+
| id   | weight | value                                   |
+------+--------+-----------------------------------------+
|    2 |      3 | Albany High School, Albany, NY          |
|    3 |      3 | South Albany High School, Albany, OR    |
|    4 |      3 | Albany High School, Albany, CA          |
|    1 |      2 | Albany Junior High School, Auckland, NZ |
+------+--------+-----------------------------------------+
4 rows in set (0.00 sec)

Changing the ranking mode does work.

回答2:

the 0's in your SetFieldWeights look odd. Either just only note the fields you want to set the weight to, or use 1 as the default. I suspect 0 will cause issues.

Suspect that SPH_RANK_SPH04 would be the most suitable for this particular case.

also shouldnt need your setSelect

来源：https://stackoverflow.com/questions/11518313/sphinx-ignores-ranking-always-sorts-in-the-same-order

标签

full-text-search

sphinx