Open Ended Question Analysis System and Method

The present invention relates to a system and method for data processing responses to open-ended questions, and, more particularly, to a system and method for providing reproducible dynamically variable coding of open-ended respondent answers to open-ended questions.

Typically, Survey questionnaires, such as used by way of example in surveying consumer response to products, normally contain three basic types of questions, i.e., closed-ended questions, which are questions to which all possible choices are shown so that these responses can be pre-coded for data processing, such as shown in FIG. 3A by way of example; open-ended questions, which are fill in answers whose response is unknown in advance and therefore cannot normally be pre-coded for data processing, such as shown in FIG. 3B by way of example; and hybrid questions which contain both types of possible responses, such as shown in FIG. 3C by way of example. In the prior art, open-ended respondent answers have normally first been processed manually to establish a coding structure that is, unfortunately, not readily reproducible and primarily operator dependent, relying on the operator's interpretative ability to find the best match or match to some guide which normally has to first be developed by partial data sampling of arriving responses which may or may not reflect the type of coding structure needed for the balance of the responses. If the codes then have to be varied, then all of the previously coded responses must be gone back to and re-coded. This procedure is quite time consuming and does not optimize the advantages of employing data processing. In an effort to overcome this, there have been prior art attempts at simple FIND systems which merely display the responses for coding, but do not employ any type of computer aided decision in arriving at the coding. This is so despite the increasing use of artificial intelligence in general, such as in prior art language analysis systems, such as shown in U.S. Pat. Nos. 4,731,021; 4,635,199; 4,586,160; 4,703,425; and 4,594,686. Moreover, none of these prior art systems known to applicants is capable of providing open-ended question response analysis and processing to provide reproducible categorized dynamically variable coding of respondent answers. These disadvantages of the prior art are overcome by the present invention.

A method and system for data processing open-ended respondent answers to open-ended questions for providing reproducible categorized dynamically variable coding of the open-ended respondent answers to the open-ended questions. The data processor has an updateable retrievable word dictionary of words stored therein, with the open-ended answers comprising words. The open-ended answers are input to the data processor and classified into corresponding word types such as keywords, modifier words, skip words, connectors, and negative words, with the combined keywords and associated modifiers forming key phrases. The input words are converted into corresponding binary coded words for providing a binary defined sentence corresponding to the input open-ended respondent answer. The binary defined sentence is scanned and any keywords and associated modifiers are extracted to create a retrievable file comprising key phrases formed from the extracted keywords and associated modifiers, and the keywords per se. The key phrases are sorted in the created file and the occurrences of the sorted key phrases are counted and duplicates eliminated in order to provide a net key phrase file. The net key phrase file is displayed to the operator who then groups the displayed net key phrases into a coding structure which is stored and can be updated or modified under control of the operator.

FIG. 1 is a diagrammatic illustration of an overall functional block diagram of the system of the present invention;

FIG. 2 is a functional logic flow diagram of the overall method of the present invention;

FIGS. 3A-3C are diagrammatic illustrations of three typical respondent answer types; i.e., closed-ended question, open-ended questions, and a hybrid question;

FIG. 4 is a diagrammatic illustration of a typical open-ended respondent answer and the method of the present invention as applied thereto;

FIG. 5 is a logic flow diagram of the word recognition phase of the method of the present invention;

FIG. 6 is a logic flow diagram of the binary phrase extraction phase of the method of the present invention;

FIG. 7 is a logic flow diagram of the sentence evaluation phase of the method of the present invention;

FIG. 8 is a logic flow diagram of the sentence validity phase of the method of the present invention;

FIG. 9 is a logic flow diagram of the key phrase extraction phase of the method of the present invention;

FIGS. 10-11 are logic flow diagrams of the share modifiers phase of the method of the present invention;

FIGS. 12-13 are a logic flow diagram of the apply database phase of the method of the present invention; and

FIG. 14 is a logic flow diagram of the save phrases phase of the method of the present invention.

Referring now to the drawings in detail, and initially to FIGS. 1-4, the presently preferred method of the present invention shall now be described. As will be explained in greater detail hereinafter, the presently preferred system and method of the present invention employs a computer, such as a Hewlett-Packard P.C. or personal computer, which has been conventionally programmed, such as in the language C to carry out the presently preferred method of the present invention, to process respondent answers to questionnaires, such as preferably open-ended responses which are fill in type answers, such as shown in FIG. 3B or FIG. 3C, which are not known in advance and therefore, cannot normally be pre-coded. Of course as shown and preferred in FIG. 2, the system and method of the present invention may also be employed if desired, to merge data collected in response to closed-end questions, such as shown in FIG. 3A, which are questions having known pre-coded response choices, or hybrid questions, such a shown in FIG. 3C, which contain both possible responses in the available choices. Thus, market research data, or other types of data collected using the aforementioned question types, can be coded and processed to provide reproducible categorized dynamically variable coding in accordance with the present invention. The respondent answers are preferably input in the form of English words and converted to binary defined sentences for coding and analysis, with the codes identifying specific types of answers within a given category of answers.

Preferably the respondent answers, such as the collected answers to open-ended questions, are input to the P.C. system 200. Along with the respondent answers, the respondent ID is also preferably entered so that preferably an index and reference is provided to the entry of any closed-end data also being entered for the respondent. The data is normally entered via a conventional keyboard associated with the P.C. which may be connected via a conventional modem, along with other P.C.'s to a central computer such as the HP 3000, for accumulative processing of the collected data if desired. The preferred system and method of the present invention uses word recognition, such as illustrated in FIG. 5, to classify the textual responses into keywords, which are keys to conveying a true meaning of the response and respondent answer; modifiers for these keywords, which are words attached to the keywords to form key phrases; skip words, which are words unimportant to conveying the meaning of the respondent answer and are thus, ignored; connector words, which are conjunctions which connect key phrases within the respondent answer so as to treat the answer as two separate responses separated by a connector; and negative words, which turn the key phrases into negative statements in the respondent answers, such as, by way of example, the word "dislike". There may also be unknown words which are words whose classification is unknown at the time of word recognition. Preferably the data processor 200 employs a stored dictionary of words having an associated ID number. The created dictionary is periodically updated as new words are added or input to the system so that recognition, preferably, only has to be done once. The open-ended textual data or respondent answers, which have been input to the system ar preferably converted into binary coded or defined sentences, which are numeric defined sentences, using the word ID numbers created for the dictionary, with synonyms or various forms of the word preferably using the same ID numbers. The binary defined sentences then, preferably, undergo phrase extraction, such as illustrated in FIG. 6, for each open-ended respondent answer using pre-defined rules to display each keyword phrase in the sentence, with each phrase including a keyword and possibly a modifier. Preferably, dynamic rule modification can be employed for phrases that do not comply with the pre-defined rules which are then learned by the computer or data processor and preferably adopted as a supplemental rule to follow whenever that phrase appears. The operator or user then uses the displayed keyword phrases to create net groups of keyword phrases which are coded. Preferably, these net groupings can be dynamically varied or modified at any time since all phrases are preferably stored in the data processor 200 in the created file and can readily be regrouped and automatically renumbered for every respondent ID.

The above presently preferred method is illustrated generally in the flow diagram of FIG. 2, and in more detail, in the detailed flow diagram of FIGS. 5-14 which illustrate the various preferred phases of the method of the present invention from word recognition through saving the extracted and processed phrases. These flow diagrams are self-explanatory and need not be further described. With respect to FIG. 2, as shown and preferred, documents 100, such as questionnaires, are used to collect the data for open-ended data entry 102 as well as for any closed-ended data entry 104. In order to enter the open-ended responses, the sentence words are keyed in 106 and, as mentioned above, are classified 108, into keywords, modifiers, skip words, connector and negative words, and a word dictionary 110 is created and updated. The word sentence is then converted into a binary defined sentence 112. The binary defined sentence is preferably scanned and keywords and associated modifiers are extracted 114. A file is then created 116 of the extracted keywords and of extracted key phrases. This key phrase file 116 is then preferably sorted 118 and the number of occurrences of particular key phrases is counted and duplicates eliminated 120. As mentioned above, the net key phrase file is displayed 122 and the operator after observing this display, groups the displayed net key phrase file into a coding structure 124. The coding of the open-ended responses is then updated 126. The file is updated 128 and respondent IDs may be matched 130 and the coded open-ended responses merged 132, if desired, with the coded closed-end answers or, alternatively, may be kept in a separate file from the coded closed-end answers.

An example of a set of presently preferred rules to be used in practicing the preferred method of the present invention is as follows, using combinations of words which can be found in various typical sentences, such as the following three sentence as an example:

FIG. 4 illustrates another example of a typical sentence coded in accordance with the system of the present invention, with S referring to a skip word, N to a negative word, C to a connector, K to keywords and M to a modifier. Other examples of typical analyses of sentences in accordance with the present invention are given below in Tables A-G which represent seven net groupings including taste/flavor/crispness, calories/fat/weight, price/costs, all other responses appearance/color, availability net, and health responses, respectively.

A typical word dictionary, by word type, in accordance with the present invention is also given below in Tables H-M.

As was previously mentioned, preferably synonyms are given the same work ID in the dictionary. An example of such synonym usage is given below in Table N.

By utilizing the above method and system of the present invention, open-ended respondent answers to open-ended questions may be processed in a data processor to provide reproducible categorized dynamically variable coding of the open-ended responses.