Table extraction – the key to more efficiency in your company

In today’s world, tables in all their forms and peculiarities have become an indispensable part of the everyday life of public authorities, companies and even private individuals. Like hardly any other means, they make it easier to bring clarity to data volumes and organizational information.

Tables are used in the respective document to present values or items clearly. The reader should therefore be relieved of the circumstance of having to laboriously read through continuous text in order to obtain the required information – regardless of whether it is the listing of the individual goods in an order, delivery bills or administrative information such as closing accounts or employee details.

So far, so good. But as data in pure print format, such tables are of little use to you today in view of the possibilities that digitalized data processing offers your company. For example, if you want to calculate statistics, evaluations or forecasts for specific sectors of your company from the data listed in the tables, in order

• to offer even better services to your customers,
  
• to remain innovative as a company,
  
• to be better than the competition,
  

it is essential to convert them efficiently into a machine-readable format.

Machine-readable means: not just scanning the document as a PDF or image file, but also making the text and thus the values in the document usable for intelligent machines and analysis applications.

This can include: Quantity data, sequence numbers, percentage data, article numbers, article descriptions, color data of articles, customer numbers, currency data, dates, postal codes, etc. From such data, it is possible to calculate, for example, which items were purchased particularly frequently at which times, in which postal code areas your company may have more customers than in others (and in which areas there is corresponding untapped potential for new customers), which items are frequently purchased together, and much more. The machine-readable digitization of documents is therefore not an end goal. Rather, it forms the gateway to the unlimited possibilities that advanced data analysis offers for increasing your company’s success and optimizing your business processes.

The technical bottleneck: digitizing tables efficiently

However, this is precisely where a major technical challenge becomes apparent. As practical and helpful as tables are for presentation and readability, they are just as difficult when it comes to digitizing the documents concerned in such a way that information can also be processed automatically. The individual formatting makes tables a persistent source of errors for text recognition programs that otherwise get along well with pure continuous text.

Tables are always visually integrated in documents in a certain way. For the well-read human observer, it is usually obvious at first glance where the continuous text ends and the table begins. It is different with programs: It is often not so easy for them to recognize the exact zones of the table and to assign sections correctly.

So far, there is no uniform formula that can be used to unambiguously teach a machine system how to recognize the field boundaries of a table. In each document, tables have different formats (width, height, white space, line density, etc.), which must be individually communicated to a readout program. By the way, this also applies to electronic files where a table is stored in image format, for example.

Thus, it often happens that the machine places a field boundary incorrectly, so that in the digitized output text words from two areas of the document are thrown together, which actually do not belong together in terms of content.

If this only happens with a single-page document, the resulting word salad can be subsequently corrected by human editors. However, when your company is faced with the task of converting entire file cabinets into a machine-readable format, such inaccuracies can quickly become a huge frustration factor.

Not only do they ensure that a lot of working time that could be used for more sensible things goes into troubleshooting. With every additional manual correction that is made, the risk of errors being overlooked due to human inattentiveness increases.

Further problems with the automatic extraction of tables in text documents are caused by the fact that the system sometimes mistakes parts of the continuous text for tables and then reads them in as such. In addition, special characters or dashes, which have an ordering function within the table, are repeatedly misinterpreted during automatic text recognition. By the way, the same applies to logos or other areas of the document that are “meant” as an image but can be incorrectly assigned to a table format by a machine.

As a result, reading out tables in particular becomes a particular hurdle when it comes to preparing documents for further digital processing. In plain language, this means an enormous amount of time and high personnel costs for your company, since the employees in your company ultimately have to transfer most of the values from the printed table by hand and then also thoroughly check and correct them afterwards.

DOC² – our solution specialized in table reading

To help your company efficiently digitize data from printed spreadsheet documents, we use DOC². DOC² is an artificial intelligence-based tool that intelligently extracts content from documents.

The biggest advantage is that it requires significantly less human supervision than other comparable systems. A scarce initial effort, in which the fields of a table document are marked, is sufficient to get the automatic extraction running.

If further feedback is then entered by your responsible employee during the digitization process, the system learns to implement these instructions in sequence.

The great benefits for you:

• Time saving
• Relieve your employees from tedious and monotonous digitization procedures
• Reliable data quality through high-level readout technologies

The entire process of digitizing your documents is thus significantly streamlined and your employees only need to perform minor manual processes, while the main part of the digitization process is handled automatically by the system.

The difference is roughly comparable to whether the construction workers on a building site have to heave the bricks into the wheelbarrow under their own power and then push the barrow across the site themselves, or whether they sit in the control cabin of a crane and have the crane move large quantities of building material by means of a few control movements of one hand.

In both cases, human input cannot be dispensed with. However, while in the first case the work process is extremely strenuous and time-consuming, in the second case a maximum of work performance is achieved with a minimum of human effort.

An insight into how to use our table extraction

Practically, the process of table extraction with Polydocs looks something like this for your employees:

The document is first scanned and is thus available for editing using DOC². In the editing mode for table extraction, your employee can only open the scan and, with a few clicks and mouse movements, define the exact fields and areas that are to be read out. In addition, your employee can determine exactly in which target column the values are to be sorted. The special feature of assigning custom columns (i.e. additional columns that can be formatted and edited completely individually) makes the work process especially easy – without any regex or other programming.

Cumbersome additional mouse movements or keyboard shortcuts for copying and pasting the values are eliminated and allow your employee to work through the document smoothly and quickly. Selected values are intelligently recognized by the program and automatically transferred to the desired fields as machine-readable values.

A selection of clearly arranged setting options and additional functions also allows you to design the table readout with just a few clicks in the way your employee thinks is necessary for the document in question.

This makes Polydocs the ideal tool for digitizing even complex spreadsheets with just a few simple steps and minimal supervision by your staff.

Furthermore, you do not need any special, cumbersome transmission channels for the transmission of the read and processed data. The data can be transmitted easily by e-mail or by automated assignment to a specified folder.

You want to get a visual impression of how the application of our program looks concretely? You can find a detailed tutorial with a demonstration in real time here in the video on our YouTube channel.

If you have any questions or would like to inquire about a customized solution for your application, please do not hesitate to contact us.