Questions tagged with Text processing & Analytics

Hello, I am using the sdk for java. Particularly the AnalyzeExpenses service. The result that I obtained in all my tests almost does not contain standard labels, that is, they have type "OTHER". In the documentation they mention tags like TOTAL, RECEIVER_ADDRESS, INVOICE_RECEIPT_DATE. I am only getting VENDOR_NAME and EXPENSE_ROW. I didn't find implementation examples in java to see if I'm doing something wrong. For example I have labels like "Total", "Date:", "Quantity" that are not being returned mapped to the standard. From already thank you very much

Hi, I had tried running a custom classification training job in Comprehend with less than 25 samples and 2 classes in total. The training took a lot of time , around 30 mins. Any idea on why the training is taking so long. Is there any way to increase the performance for training ?

I am trying to send a query with the function analyze_document (or start_document_analysis) as described in the docs , however I get the error: `Unknown parameter in input: "QueriesConfig", must be one of: Document, FeatureTypes, HumanLoopConfig ` If I then remove "QueriesConfig", I get the error : `An error occurred (InvalidParameterException) when calling the AnalyzeDocument operation: QueriesConfig must be used with QUERIES FeatureType. ` I do not know hence what to do. I leave the full script code. ``` client = boto3.client('textract') image_binary = stream.getvalue() response = client.analyze_document(Document={'Bytes': image_binary}, FeatureTypes=["QUERIES"], QueriesConfig={ 'Queries': [ { 'Text': 'what is the document date?' } ] }) ``` Thanks!

I have a csv of sentences which I'd like labelled, and have identified GroundTruth labelling jobs as a way to do this. Having spent some time exploring the service, I have some questions: **1) **I can't find a way to display only particular columns to the labellers - e.g. if the dataset has a column of IDs for each sentence, this ideally shouldn't be shown to labellers **2)** There is either single labelling or multi labelling, but I would like a way to have two sets of single-selection labels, where one captures difficulty of assigning the label: Select one for binary classification a) Yes, b) No Select one for difficulty of classification c) Easy, d) Medium, e) Hard Can this be done using custom HTML? Is there a guide to writing this - the template it gives you doesn't seem to render as-is. **3)** There appears to be a maximum of $1.20 payment per task. Is this the case, and why? **4)** Having not used mechanical turk before, are there ways of ensuring people take the work seriously and don't just select random answers? I can see there's an option to have x number of people answer the same question, but is there also a way to put in unambiguous questions to which we already have a 'pre_agreed_label' every nth question, and remove people from the task if they get them wrong? Thanks!