How to parse a HTML table with Nokogiri?

The key of the problem is that calling #text on multiple results will return the concatenation of the #text of each individual element.

Lets examine what each step does:

# Finds all <table>s with class open
# I'm assuming you have only one <table> so
#  you don't actually have to loop through
#  all tables, instead you can just operate
#  on the first one. If that is not the case,
#  you can use a loop the way you did
tables = doc.css('table.open')

# The text of all <th>s in <tr> one in the table
title = table.css('tr[1] > th').text

# The text of all <td>s in all <tr>s in the table
# You obviously wanted just the <td>s in one <tr>
cell_data = table.css('tr > td').text

# The text of all <th>s in all <tr>s in the table
# You obviously wanted just the <th>s in one <tr>
raw_name = table.css('tr > th').text

Now that we know what is wrong, here is a possible solution:

html = <<EOT
    <table class="open">
        <tr>
            <th>Table name</th>
            <th>Column name 1</th>
            <th>Column name 2</th>
            <th>Column name 3</th>
            <th>Column name 4</th>
            <th>Column name 5</th>
        </tr>
        <tr>
            <th>Raw name 1</th>
            <td>1001</td>
            <td>1002</td>
            <td>1003</td>
            <td>1004</td>
            <td>1005</td>         
        </tr>
        <tr>
            <th>Raw name 2</th>
            <td>2001</td>
            <td>2002</td>
            <td>2003</td>
            <td>2004</td>
            <td>2005</td>         
        </tr>
        <tr>
            <th>Raw name 3</th>
            <td>3001</td>
            <td>3002</td>
            <td>3003</td>
            <td>3004</td>
            <td>3005</td>         
        </tr>
    </table>
EOT

doc = Nokogiri::HTML(html, nil, 'UTF-8')

# Fetches only the first <table>. If you have
#  more than one, you can loop the way you
#  originally did.
table = doc.css('table.open').first

# Fetches all rows (<tr>s)
rows = table.css('tr')

# The column names are the first row (shift returns
#  the first element and removes it from the array).
# On that row we get the text of each individual <th>
# This will be Table name, Column name 1, Column name 2...
column_names = rows.shift.css('th').map(&:text)

# On each of the remaining rows
text_all_rows = rows.map do |row|

  # We get the name (<th>)
  # On the first row this will be Raw name 1
  #  on the second - Raw name 2, etc.
  row_name = row.css('th').text

  # We get the text of each individual value (<td>)
  # On the first row this will be 1001, 1002, 1003...
  #  on the second - 2001, 2002, 2003... etc
  row_values = row.css('td').map(&:text)

  # We map the name, followed by all the values
  [row_name, *row_values]
end

p column_names  # => ["Table name", "Column name 1", "Column name 2",
                #     "Column name 3", "Column name 4", "Column name 5"]
p text_all_rows # => [["Raw name 1", "1001", "1002", "1003", "1004", "1005"],
                #     ["Raw name 2", "2001", "2002", "2003", "2004", "2005"],
                #     ["Raw name 3", "3001", "3002", "3003", "3004", "3005"]]

# If you want to combine them
text_all_rows.each do |row_as_text|
  p column_names.zip(row_as_text).to_h
end # =>
    # {"Table name"=>"Raw name 1", "Column name 1"=>"1001", "Column name 2"=>"1002", "Column name 3"=>"1003", "Column name 4"=>"1004", "Column name 5"=>"1005"}
    # {"Table name"=>"Raw name 2", "Column name 1"=>"2001", "Column name 2"=>"2002", "Column name 3"=>"2003", "Column name 4"=>"2004", "Column name 5"=>"2005"}
    # {"Table name"=>"Raw name 3", "Column name 1"=>"3001", "Column name 2"=>"3002", "Column name 3"=>"3003", "Column name 4"=>"3004", "Column name 5"=>"3005"}

Your desired output is nonsense:

['Raw name 1', 2,094, 0,017, 0,098, 0,113, 0,452]
# ~> -:1: Invalid octal digit
# ~> ['Raw name 1', 2,094, 0,017, 0,098, 0,113, 0,452]

I'll assume you want quoted numbers.

After stripping the stuff that keeps the code from working, and reducing the HTML to a more manageable example, then running it:

require 'nokogiri'

html = <<EOT
    <table class="open">
        <tr>
            <th>Table name</th>
            <th>Column name 1</th>
            <th>Column name 2</th>
        </tr>
        <tr>
            <th>Raw name 1</th>
            <td>2,094</td>
            <td>0,017</td>
        </tr>
        <tr>
            <th>Raw name 5</th>
            <td>2,094</td>
            <td>0,017</td>
        </tr>
    </table>
EOT


doc = Nokogiri::HTML(html)
tables = doc.css('table.open')

tables_data = []

tables.each do |table|
  title = table.css('tr[1] > th').text # !> assigned but unused variable - title
  cell_data = table.css('tr > td').text
  raw_name = table.css('tr > th').text
  tables_data << [cell_data, raw_name]
end

Which results in:

tables_data
# => [["2,0940,0172,0940,017",
#      "Table nameColumn name 1Column name 2Raw name 1Raw name 5"]]

The first thing to notice is you're not using title though you assign to it. Possibly that happened when you were cleaning up your code as an example.

css, like search and xpath, returns a NodeSet, which is akin to an array of Nodes. When you use text or inner_text on a NodeSet it returns the text of each node concatenated into a single string:

Get the inner text of all contained Node objects.

This is its behavior:

require 'nokogiri'

doc = Nokogiri::HTML('<html><body><p>foo</p><p>bar</p></body></html>')

doc.css('p').text # => "foobar"

Instead, you should iterate over each node found, and extract its text individually. This is covered many times here on SO:

doc.css('p').map{ |node| node.text } # => ["foo", "bar"]

That can be reduced to:

doc.css('p').map(&:text) # => ["foo", "bar"]

See "How to avoid joining all text from Nodes when scraping" also.

The docs say this about content, text and inner_text when used with a Node:

Returns the content for this Node.

Instead, you need to go after the individual node's text:

require 'nokogiri'

html = <<EOT
    <table class="open">
        <tr>
            <th>Table name</th>
            <th>Column name 1</th>
            <th>Column name 2</th>
            <th>Column name 3</th>
            <th>Column name 4</th>
            <th>Column name 5</th>
        </tr>
        <tr>
            <th>Raw name 1</th>
            <td>2,094</td>
            <td>0,017</td>
            <td>0,098</td>
            <td>0,113</td>
            <td>0,452</td>         
        </tr>
        <tr>
            <th>Raw name 5</th>
            <td>2,094</td>
            <td>0,017</td>
            <td>0,098</td>
            <td>0,113</td>
            <td>0,452</td>         
        </tr>
    </table>
EOT


tables_data = []

doc = Nokogiri::HTML(html)

doc.css('table.open').each do |table|

  # find all rows in the current table, then iterate over the second all the way to the final one...
  table.css('tr')[1..-1].each do |tr|

    # collect the cell data and raw names from the remaining rows' cells...
    raw_name = tr.at('th').text
    cell_data = tr.css('td').map(&:text)

    # aggregate it...
    tables_data += [raw_name, cell_data]
  end
end

Which now results in:

tables_data
# => ["Raw name 1",
#     ["2,094", "0,017", "0,098", "0,113", "0,452"],
#     "Raw name 5",
#     ["2,094", "0,017", "0,098", "0,113", "0,452"]]

You can figure out how to coerce the quoted numbers into decimals acceptable to Ruby, or manipulate the inner arrays however you want.