Dialog management system
A dialog management system functioned by a special search algorithm, the search algorithm comprises the rule of recognizing the phrase directly, converting the words into terms, filtering the adjacency terms and recognizing, the dynamic field the Dialog Management System comprises: an application conducting a predefined operation between a user and a terminal; an input unit receiving an utterance from the user; recognition unit recognizing and converting the received utterance from sound into sequential words in a written text; processor; wherein the processor comprises dynamic field recognition unit; conversion unit, filter unit, a matching unit; a database storing a dictionary that includes the all of the definition of words and the word-term list.
1. A computer implemented method for dialog management, comprising the steps of:
S 1000 : receiving an utterance from a user by a terminal;
S 2000 : recognizing and converting the utterance from sound into sequential words in a written text;
S 3000 : determining if there is a phrase in the written text matching with a predefined phrase in a menu tree; wherein the menu tree comprises a root node, the root node comprises a plurality of parent nodes, and each parent node comprises at least one leaf node, each leaf node corresponds to an operation, each of the predefined phrases corresponds to a parent node or a leaf node;
if the phrase in the written text matches with more than one of the predefined phrases in the menu tree,
S 3110 : attempting to recognize at least one dynamic field in the written text;
S 3120 : converting the sequential words into corresponding predefined terms in sequence, wherein different words with similar meaning are converted into a term; each of the predefined terms corresponds to a parent node or a leaf node;
S 3130 : matching terms with the corresponding parent node or leaf node in a matched phrase node;
S 4000 : conducting the operation according to the matched phrase node;
if the phrase in the written text matches with exactly one of the predefined phrase in the menu tree,
S 3210 : attempting to recognize at least one dynamic field in the written text;
S 3220 : matching the phrases with the parent node or the leaf node,
S 4000 : conducting the operation according to the matched phrase node;
if the phrase in the written text does not matches with any of the predefined phrase in the menu tree,
S 3310 : attempting to recognize at least one dynamic field in the written text;
S 3320 : converting the sequential words into corresponding predefined terms in sequence, wherein different words indicating the similar intent are converted into a term;
S 3330 : matching the terms with the corresponding parent node or leaf node in all nodes;
S 4000 : conducting the operation according to the matched phrase node.
2. The method according to claim 1 , wherein said dynamic field is used for intent matching.
3. The method according to claim 1 , wherein a minimum and a maximum occurrence threshold are predetermined for recognizing the dynamic field in order to handle inadequate or excessive data.
4. The method according to claim 3 , further comprising:
executing an indicator of a result of the recognized dynamic fields, in order to decide whether all necessary dynamic fields are collected or not;
if the necessary dynamic fields have not been recognized, prompting to ignore the missing dynamic fields or starting another recognition.
5. The method according to claim 1 , further comprising:
after the step of converting the sequential words into corresponding predefined terms in sequence, filtering the converted terms to obtain key terms.
6. The method according to claim 5 , wherein the step of filtering the converted terms to obtain key terms is carried out by a predefined adjacency filter rule.
7. The method according to claim 1 , wherein,
the computer is capable of announcing a given text or a given record to the user according to the conducted operation.
8. The method according to claim 7 , wherein the announced text or record comprises general announcements and specific announcements; wherein the computer is capable of fusing the general announcements and the specific announcements.
9. The method according to claim 7 , wherein, if there is multiple matching, the computer is capable of creating specific multiple match responses.
10. The method according to claim 7 , wherein the dynamic fields are used to create multiple announcements with similar intent.
11. The method according to claim 7 , wherein an announcement database is created for a plurality of customer segments; wherein the plurality of customer segments are categorized according to age, education, and demography of customers; wherein for the same conducted operation, the announcement corresponding to the customer segment in which the user falls is played from the announcement database.
12. The method according to claim 11 , wherein the announcement database corresponding to each customer segment is further divided into a plurality of sub-databases corresponding to a plurality of system personas, wherein each sub-database in a particular customer segment provides a different announcement for the same conducted operation, wherein the announcement for the same conducted operation is selected from a specific sub-database based on a predefined circumstance.
13. The method according to claim 1 , wherein sentences containing the dynamic fields are given to the computer in order to produce training sentences.
14. The method according to claim 1 , wherein a plurality of responses are assigned for each operation and the computer randomly selects and announces a first response from said plurality of responses for the same conducted operation; after announcing said one response if the user inputs the same operation again, the computer randomly selects and announces a second response from said plurality of responses, wherein the second response is different from the first response.