[textplot_dependencyparser.default] Add layout argument

[discoleo]

Function textplot_dependencyparser.default

Add a layout argument:

• default value: layout = "linear";

Code changes

[...]
ggraph::ggraph(g, layout = layout) +
[...]

Example

Problem:

• the plot fails ugly with very long sentences.;
• example uses the Results section of an abstract from Pubmed (PMID = 35420756);
• I have uploaded a set of 1200 Pubmed abstracts on Github, see file Example_Abstracts_Title_Pubmed.csv

  
  abs1 = "RESULTS: Our scan of Canadian government and university web pages found that  
        universities similarly created one main page for COVID-19 updates and information
        and linked to public sector agencies as a main resource, and mainly differed in  
        their provincial and local sources for obtaining information. They also differed 
        in their strategies for communicating and displaying this information to their   
        respective students."
  txtSection = "(?i)^(?:Objectives|Results|Conclusions)\\:"
  tmp = sub(txtSection, "", abs1);
  # udmodel = english-ewt;
  ann = udpipe_annotate(udmodel, x = tmp)
  ann = as.data.frame(ann, detailed = TRUE)

Sentence

idS = 1 textplot_dependencyparser(ann[ann$sentence_id == idS,])

[discoleo]

Basic code for Multi-line Layout

Note:

• the split may not be optimal;
• TODO: fix (remaining) width and text-overlap issues;
• slightly improved & fixed version of the function;
• default: 2 rows (nr = 2);

layoutSplit = function(x, y=NULL, nr=2, scale=c(1, 30), dx=3) {
    len = nchar(x);
    if( ! is.null(y)) len = pmax(len, nchar(y));
    len = len + dx;
    if(nr > 1) {
        nm = length(len) %% nr;
        if(nm != 0) len = c(len, rep(0, nr - nm));
    } else nm = 0;
    nc  = length(len) %/% nr;
    len = matrix(len, ncol=nr);
    # x-Coord = cumsum() + mid;
    pos = apply(len, 2, cumsum);
    pos = rbind(0, pos[ - nrow(pos), ]);
    pos = pos + len / 2; # midpoint
    dim(pos) = NULL;
    pos = pos * scale[1];
    xdf = data.frame(
        x = pos,
        y = rep(seq(nr, 1, by=-1) * scale[2], each=nc));
    if(nm > 0) {
        xdf = xdf[ seq(nr - nm) - 1 - nrow(xdf), ];
    }
    return(xdf)
}

Usage

• improved labeling requires that the nudge_y argument is also visible;

  idS  = 1
  txtS = ann[ann$sentence_id == idS, ];
  textplot_dependencyparser(txtS, layout = layoutSplit(txtS$token, txtS$upos, nr=2), nudge_y = -5)

• proof of concept hack (based on internal code);

  [...]
  ggraph::ggraph(g, layout = layoutSplit(x$token, y = x$upos, nr=2)) +
  [...]

[jwijffels]

We can certainly add the layout argument in the function definition so that you can pass whichever on.

Note getting this currently with your layoutSplit

[discoleo]

Latest version of layoutSplit:

• fixed a bug & improved layout;

Note:

• the function splits the number of words into nr rows;
• it does not split the number of chars evenly, nor does it optimize for head tokens;

[jwijffels]

I've added the layout argument in commit https://github.com/bnosac/textplot/commit/6097d2b0453ad19d01b03e55dd17e88bccbbec8d and will upload the package to CRAN