Identifying section headings in a document

1. A method for processing an electronic document (ED) to infer a sequence of section headings in the ED, the method comprising:

generating, by a computer processor and based on a statistical distribution of point sizes of a plurality of characters in the ED, a list of candidate headings in the ED, each candidate heading corresponding to a paragraph in the ED, wherein

the list of candidate headings comprises a plurality of same-point-size groups,

each of the same-point-size groups correspond to a unique point size, and

the ED comprises at least three candidate headings;

adding, by the computer processor, an initial chain fragment from the list of candidate headings to initialize a style chain, wherein the initial chain fragment is removed from the list of candidate headings subsequent to the adding;

iteratively augmenting, based on an iterative style chain generation algorithm, the style chain to include a plurality of chain fragments, comprising:

generating, during a first iteration among a plurality of iterations of the iterative style chain generation algorithm, a first chain fragment among the plurality of chain fragments from a first same-point-size group among the plurality of same-point-size groups, wherein

the first chain fragment comprises a parent adjacent candidate heading pair (ACHP) for a second iteration among the plurality of iterations, and

the first chain fragment is added to the style chain and removed from the list of candidate headings; and

generating, during the second iteration subsequent to the first iteration, a second chain fragment among the plurality of chain fragments from a second same-point-size group among the plurality of same-point-size groups, wherein

the second chain fragment is bounded by the parent ACHP for the second iteration,

the second chain fragment is added to the style chain and removed from the list of candidate headings,

the unique point size of the second same-point-size group is less than or equal to the unique point size of the first same-point-size group, and

the style chain comprises at least two but not all candidate headings; and

generating, by the computer processor and based at least on the plurality of chain fragments of the style chain, the sequence of section headings.

2. The method of claim 1 , wherein

generating the first chain fragment during the first iteration comprises:

selecting a first current position and a first end bound from a first plurality of paragraph numbers in the style chain;

selecting, based on the first current position and the first end bound, the first chain fragment from the list of candidate headings, and

removing, prior to the second iteration and in response to selecting the first chain fragment, at least the first chain fragment from the list of candidate headings to generate a remaining list of candidate headings; and

generating the second chain fragment during the second iteration comprises:

selecting a second current position and a second end bound from a second plurality of paragraph numbers in the style chain;

selecting, based on the second current position and the second end bound, the second chain fragment from the remaining list of candidate headings.

3. The method of claim 2 , wherein generating the second chain fragment during the second iteration further comprises:

identifying an alternative chain fragment based on an alternative same-point-size group from the plurality of same-point-size groups, wherein

the alternative chain fragment is bounded by the parent ACHP in the first chain fragment, and

the unique point size of the alternative same-point-size group is less than or equal to the unique point size of the first same-point-size group;

comparing a measure of children generation potential between the second chain fragment and the alternative chain fragment to generate a comparison result; and

discarding the alternative chain fragment based on the comparison result.

4. The method of claim 2 , where selecting the second current position and the second end bound comprises:

identifying the second current position and the second end bound based on a leading paragraph number and a trailing paragraph number of the parent ACHP for the second iteration.

5. The method of claim 4 , wherein

the second iteration follows the first iteration without any intervening iteration,

the second current position equals the leading paragraph number of the parent ACHP for the second iteration, and

the second end bound equals one less than the trailing paragraph number of the parent ACHP for the second iteration.

6. The method of claim 4 , wherein

the second iteration is preceded by an intervening iteration that follows the first iteration and results in no additional chain fragment, the intervening iteration being based on an intervening current position and an intervening end bound,

the second current position equals one plus the intervening end bound, and

the second end bound equals one less than the trailing paragraph number of the parent ACHP for the second iteration.

7. The method of claim 1 , further comprising:

identifying, in the ED, one or more explicit section heading, a plurality of stylized candidate headings, and a plurality of sequenced candidate headings, wherein

the style chain comprises the one or more explicit section heading prior to adding the initial chain fragment to the style chain, and

the list of candidate headings comprises the plurality of stylized candidate headings and the plurality of sequenced candidate headings.

8. The method of claim 1 , further comprising:

identifying a main title of the ED based on the sequence of section heading.

9. A non-transitory computer readable medium (CRM) storing computer readable program code for processing an electronic document (ED) to infer a sequence of section headings in the ED embodied therein, wherein the computer readable program code, when executed by a computer, comprises functionality for:

generating, based on a statistical distribution of point sizes of a plurality of characters in the ED, a list of candidate headings in the ED, each candidate heading corresponding to a paragraph in the ED, wherein

the list of candidate headings comprises a plurality of same-point-size groups,

each of the same-point-size groups corresponding to a unique point size, and

the ED comprises at least three candidate headings;

adding, an initial chain fragment from the list of candidate headings to initialize a style chain, wherein the initial chain fragment is removed from the list of candidate headings subsequent to the adding;

iteratively augmenting, based on an iterative style chain generation algorithm, the style chain to include a plurality of chain fragments, comprising:

generating, during a first iteration among a plurality of iterations of the iterative style chain generation algorithm, a first chain fragment among the plurality of chain fragments from a first same-point-size group among the plurality of same-point-size groups, wherein

the first chain fragment comprises a parent adjacent candidate heading pair (ACHP) for a second iteration among the plurality of iterations,

the first chain fragment is added to the style chain and removed from the list of candidate headings; and

generating, during the second iteration subsequent to the first iteration, a second chain fragment among the plurality of chain fragments from a second same-point-size group among the plurality of same-point-size groups, wherein

the second chain fragment is bounded by the parent ACHP for the second iteration,

the second chain fragment is added to the style chain and removed from the list of candidate headings,

the unique point size of the second same-point-size group is less than or equal to the unique point size of the first same-point-size group, and

the style chain comprises at least two but not all candidate headings; and

generating, based at least on the plurality of chain fragments of the style chain, the sequence of section headings.

10. The non-transitory CRM of claim 9 , wherein

generating the first chain fragment during the first iteration comprises:

selecting a first current position and a first end bound from a first plurality of paragraph numbers in the style chain;

selecting, based on the first current position and the first end bound, the first chain fragment from the list of candidate headings, and

removing, prior to the second iteration and in response to selecting the first chain fragment, at least the first chain fragment from the list of candidate headings to generate a remaining list of candidate headings; and

generating the second chain fragment during the second iteration comprises:

selecting a second current position and a second end bound from a second plurality of paragraph numbers in the style chain;

selecting, based on the second current position and the second end bound, the second chain fragment from the remaining list of candidate headings.

11. The non-transitory CRM of claim 10 , wherein generating the second chain fragment during the second iteration further comprises:

identifying an alternative chain fragment based on an alternative same-point-size group from the plurality of same-point-size groups, wherein

the alternative chain fragment is bounded by the parent ACHP in the first chain fragment, and

the unique point size of the alternative same-point-size group is less than or equal to the unique point size of the first same-point-size group;

comparing a measure of children generation potential between the second chain fragment and the alternative chain fragment to generate a comparison result; and

discarding the alternative chain fragment based on the comparison result.

12. The non-transitory CRM of claim 10 , where selecting the second current position and the second end bound comprises:

identifying the second current position and the second end bound based on a leading paragraph number and a trailing paragraph number of the parent ACHP for the second iteration.

13. The non-transitory CRM of claim 12 , wherein

the second iteration follows the first iteration without any intervening iteration,

the second current position equals the leading paragraph number of the parent ACHP for the second iteration, and

the second end bound equals one less than the trailing paragraph number of the parent ACHP for the second iteration.

14. The non-transitory CRM of claim 12 , wherein

the second iteration is preceded by an intervening iteration that follows the first iteration and results in no additional chain fragment, the intervening iteration being based on an intervening current position and an intervening end bound,

the second current position equals one plus the intervening end bound, and

the second end bound equals one less than the trailing paragraph number of the parent ACHP for the second iteration.

15. A system for processing an electronic document (ED) to infer a sequence of section headings in the ED, the system comprising:

a memory; and

a computer processor connected to the memory and configured to:

generate, based on a statistical distribution of point sizes of a plurality of characters in the ED, a list of candidate headings in the ED, each candidate heading corresponding to a paragraph in the ED, wherein

the list of candidate headings comprises a plurality of same-point-size groups,

each of the same-point-size groups correspond to a unique point size, and

the ED comprises at least three candidate headings;

add an initial chain fragment from the list of candidate headings to initialize a style chain, wherein the initial chain fragment is removed from the list of candidate headings subsequent to the adding;

iteratively augment, based on an iterative style chain generation algorithm, the style chain to include a plurality of chain fragments, comprising:

generating, during a first iteration among a plurality of iterations of the iterative style chain generation algorithm, a first chain fragment among the plurality of chain fragments from a first same-point-size group among the plurality of same-point-size groups, wherein

the first chain fragment comprises a parent adjacent candidate heading pair (ACHP) for a second iteration among the plurality of iterations, and

the first chain fragment is added to the style chain and removed from the list of candidate headings; and

generating, during the second iteration subsequent to the first iteration, a second chain fragment among the plurality of chain fragments from a second same-point-size group among the plurality of same-point-size groups, wherein

the second chain fragment is bounded by the parent ACHP for the second iteration,

the second chain fragment is added to the style chain and removed from the list of candidate headings,

the unique point size of the second same-point-size group is less than or equal to the unique point size of the first same-point-size group, and

the style chain comprises at least two but not all candidate headings; and

generate, based at least on the plurality of chain fragments of the style chain, the sequence of section headings.

16. The system of claim 15 , wherein

generating the first chain fragment during the first iteration comprises:

selecting a first current position and a first end bound from a first plurality of paragraph numbers in the style chain;

selecting, based on the first current position and the first end bound, the first chain fragment from the list of candidate headings, and

removing, prior to the second iteration and in response to selecting the first chain fragment, at least the first chain fragment from the list of candidate headings to generate a remaining list of candidate headings; and

generating the second chain fragment during the second iteration comprises:

selecting a second current position and a second end bound from the first plurality of paragraph numbers in the style chain;

selecting, based on the second current position and the second end bound, the second chain fragment from the remaining list of candidate headings.

17. The system of claim 16 , wherein generating the second chain fragment during the second iteration further comprises:

identifying an alternative chain fragment based on an alternative same-point-size group from the plurality of same-point-size groups, wherein

the alternative chain fragment is bounded by the parent ACHP in the first chain fragment, and

the unique point size of the alternative same-point-size group is less than or equal to the unique point size of the first same-point-size group;

comparing a measure of children generation potential between the second chain fragment and the alternative chain fragment to generate a comparison result; and

discarding the alternative chain fragment based on the comparison result.