1. 技术文章
  2. solr scoring - fieldnorm

solr scoring - fieldnorm

由 巧了我就是萌 提交于 2019-12-01 19:48:15

fieldnorm takes into account the field length i.e. the number of terms.
The fieldtype used is text for the fields display name & name, which would have the stopwords and the word delimiter filters.

Record 1 - Iphone
Would generate a single token - IPhone

Record 2 - The Iphone Book
Would generate 2 tokens - Iphone, Book
The would be removed by the stopwords.

Record 3 - iPhone
Would also generate 2 tokens - i,phone
As iPhone has a case change, the word delimiter filter with splitOnCaseChange would now split iPhone into 2 tokens i, Phone and would produce the field norm same as Record 2

This is the answer to user1021590's follow-up question/answer on the "da vinci code" search example.

The reason all the documents get the same score is due to a subtle implementation detail of lengthNorm. Lucence TFIDFSimilarity doc states the following about norm(t, d):

the resulted norm value is encoded as a single byte before being stored. At search time, the norm byte value is read from the index directory and decoded back to a float norm value. This encoding/decoding, while reducing index size, comes with the price of precision loss - it is not guaranteed that decode(encode(x)) = x. For instance, decode(encode(0.89)) = 0.75.

If you dig into the code, you see that this float-to-byte encoding is implemented as follows:

public static byte floatToByte315(float f)
{
    int bits = Float.floatToRawIntBits(f);
    int smallfloat = bits >> (24 - 3);
    if (smallfloat <= ((63 - 15) << 3))
    {
        return (bits <= 0) ? (byte) 0 : (byte) 1;
    }
    if (smallfloat >= ((63 - 15) << 3) + 0x100)
    {
        return -1;
    }
    return (byte) (smallfloat - ((63 - 15) << 3));
}

and the decoding of that byte to float is done as:

public static float byte315ToFloat(byte b)
{
    if (b == 0)
        return 0.0f;
    int bits = (b & 0xff) << (24 - 3);
    bits += (63 - 15) << 24;
    return Float.intBitsToFloat(bits);
}

lengthNorm is calculated as 1 / sqrt( number of terms in field ). This is then encoded for storage using floatToByte315. For a field with 3 terms, we get:

floatToByte315( 1/sqrt(3.0) ) = 120

and for a field with 4 terms, we get:

floatToByte315( 1/sqrt(4.0) ) = 120

so both of them get decoded to:

byte315ToFloat(120) = 0.5.

The doc also states this:

The rationale supporting such lossy compression of norm values is that given the difficulty (and inaccuracy) of users to express their true information need by a query, only big differences matter.

UPDATE: As of Solr 4.10, this implementation and corresponding statements are part of DefaultSimilarity.

标签
solr
scoring
易学教程内所有资源均来自网络或用户发布的内容，如有违反法律规定的内容欢迎反馈！
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!